Continual instruction tuning (CIT) has emerged as a promising strategy for adapting large language models (LLMs) to new tasks while preserving historical knowledge. Most existing CIT methods have focused on offline CIT (offCIT), which assumes clearly defined task boundaries and allows multiple passes over the data. However, such assumptions rarely hold in real-world scenarios, where data arrive in a streaming fashion and task boundaries are unknown. This setting introduces critical challenges: the absence of task identifiers (task IDs), a significant imbalance in task-specific information, and inaccessibility to previously seen data. In this work, we propose Online Editing with Decoupled Implicit Task (OnEDIT), an online CIT(onCIT) approach to tackle these challenges. OnEDIT leverages a fixed-size adapter for the implicit task, balancing current and past knowledge through editing operations every time step without relying on task IDs or backpropagation. Extensive experiments on CIT benchmarks demonstrate that OnEDIT consistently maintains robust and stable performance, whereas existing state-of-the-art baselines often suffer from performance degradation in online settings. It suggests that OnEDIT achieves superior generalization across diverse task orders and model scales, while maintaining high efficiency and low memory overhead.

Neural transducer models learn alignments implicitly by marginalizing over all valid paths, which can lead to unstable or sub-optimal alignments under low-resource and streaming conditions where alignment ambiguity is prevalent. Previous approaches such as knowledge distillation and pruned transducers attempt to mitigate this, but they either rely on large teacher models or impose rigid pruning strategies that limit flexibility. In this work, we propose a self-guided alignment regularization framework based on optimal transport (OT), which computes a soft and globally consistent alignment plan between encoder and prediction embeddings. We further extend this to a dynamic unbalanced OT formulation with relaxed marginal constraints and integrate OT-derived alignment features into the joint network via residual fusion. Experiments on the LibriSpeech 100h and 960h benchmarks showed that our method consistently improved alignment quality and recognition accuracy, yielding relative WER reductions of 14.45%/6.27% ontest-clean/test-otherin the low-resource setting and 25.49%/10.81% in the high-resource setting, without increasing inference complexity.

Noise-robust automatic speech recognition (ASR) has been commonly addressed by applying speech enhancement (SE) at the waveform level before recognition. However, speech-level enhancement does not always translate into consistent recognition improvements due to residual distortions and mismatches with the latent space of the ASR encoder. In this letter, we introduce a complementary strategy termed latent-level enhancement, where distorted representations are refined during ASR inference. Specifically, we propose a plug-and-play Flow Matching Refinement module (FM-Refiner) that operates on the output latents of a pretrained CTC-based ASR encoder. Trained to map imperfect latents—either directly from noisy inputs or from enhanced-but-imperfect speech—toward their clean counterparts, the FM-Refiner is applied only at inference, without fine-tuning ASR parameters. Experiments show that FM-Refiner consistently reduces word error rate, both when directly applied to noisy inputs and when combined with conventional SE front-ends. These results demonstrate that latent-level refinement via flow matching provides a lightweight and effective complement to existing SE approaches for robust ASR.

We propose MoCo-SSL, a momentum-based contrastive learning framework for multi-channel sound source localization (SSL) that enhances azimuth-aware representation learning. While prior SSL studies have used contrastive learning to handle varied acoustic conditions, we emphasize hard negatives-pairs with distinct azimuths recorded in the same room-for learning fine-grained spatial cues. A curriculum-based strategy gradually increases the proportion of such samples to raise task difficulty. The momentum contrast design employs a key encoder that maintains stable embeddings during curriculum transitions and receives audio with less noise and reverberation to produce clearer azimuth cues, thereby guiding the query encoder toward robust representations. Experiments show that MoCo-SSL consistently surpasses baselines, demonstrating the value of structured and noise-resilient representation learning in challenging SSL scenarios.

Robust audio-visual speech recognition (AVSR) in noisy environments remains challenging, as existing systems struggle to estimate audio reliability and dynamically adjust modality reliance. We propose router-gated cross-modal feature fusion, a novel AVSR framework that adaptively reweights audio and visual features based on token-level acoustic corruption scores. Using an audio-visual feature fusion-based router, our method down-weights unreliable audio tokens and reinforces visual cues through gated cross-attention in each decoder layer. This enables the model to pivot toward the visual modality when audio quality deteriorates. Experiments on LRS3 demonstrate that our approach achieves an 16.51-42.67% relative reduction in word error rate compared to AV-HuBERT. Ablation studies confirm that both the router and gating mechanism contribute to improved robustness under real-world acoustic noise.

Audio signal processing systems often operate differently depending on the recording devices, leading to performance discrepancies. Therefore, it is important to know about the characteristics of the recording device; however, it is difficult to know the device’s behavior in most cases. In this study, we propose a diffusion-model-based device characteristic transfer to estimate the device’s frequency response only with the recorded signals. By joint-training the conditional and unconditional diffusion models, it is found that non-linear distortions and some filtered signals are reflected more than by only training the conditional model. We show that the proposed method transfers the style closely to the ground truth not only visually on the spectrogram but also the t-distributed stochastic neighbor embedding distribution and the performance of the device classifier. We also show the proposed method enhancing the performance as a data augmentation method for acoustic scene classification.

Adaptive S-norm (AS-norm) calibrates automatic speaker verification (ASV) scores by normalizing them utilize the scores of impostors which are similar to the input speaker. However, AS-norm does not involve any learning process, limiting its ability to provide appropriate regularization strength for various evaluation utterances. To address this limitation, we propose a trainable AS-norm (TAS-norm) that leverages learnable impostor embeddings (LIEs), which are used to compose the cohort. These LIEs are initialized to represent each speaker in a training dataset consisting of impostor speakers. Subsequently, LIEs are fine-tuned by simulating an ASV evaluation. We utilize a margin penalty during top-scoring IEs selection in fine-tuning to prevent non-impostor speakers from being selected. In our experiments with ECAPA-TDNN, the proposed TAS-norm observed 4.11% and 10.62% relative improvement in equal error rate and minimum detection cost function, respectively, on VoxCeleb1-O trial compared with standard AS-norm without using proposed LIEs. We further validated the effectiveness of the TAS-norm on additional ASV datasets comprising Persian and Chinese, demonstrating its robustness across different languages.

Effectively recognizing emotions requires sophisticated approaches for interpreting diverse modalities, particularly in real-world scenarios where multiple data sources, such as speech, text, and visual cues, are often noisy and incomplete. This study proposes an advanced multimodal emotion recognition system that integrates these three modalities by adding the speaker detection and extraction algorithm within visual data. The pre-trained Q-Former used in the proposed system then captures and interprets visual signals supported with designated prompts, resulting in facial-related features that significantly improve emotion recognition performance. We then utilize a cross-modal transformer to unify the visual, speech, and text embeddings for accurate emotion classification. We achieved a 2.9% and 3.3% improvement in accuracy and F1 score, respectively, on the MELD dataset compared to the baseline.

In this paper, we consider new reconstruction loss together with a subband objective in the form of auxiliary loss function for artifacts-free speech super-resolution. Unlike prior work which mainly consider full band of frequency region for speech super-resolution, the proposed method alleviates distortion generated during deep learning training via subband modeling. To further minimize spectral artifacts, we also apply progressive curriculum learning for superior performance. Our experimental results demonstrate that the proposed method outperforms the evaluated baselines on the both TIMIT and VCTK dataset by increase in both intelligibility and perceptual score. Furthermore, the visual representation of spectrograms comparison verify that our proposed method clearly restoring speech with fewer artifacts. Audio samples and the implementations are available online.1

Recognizing not only pre-defined keywords but also continuously expanding new keywords often with limited data has emerged as a main problem in recent keyword spotting research. To address this challenge few-shot class-incremental learning approaches have gained attention initially training models on sufficient data in a base session and then continuously adapting to recognize new classes with limited data. Recent focus has been on prototype-based calibration which fuses new prototypes with weighted base prototypes. However this method risks misclassification due to increased similarity between new and base classes. To mitigate this issue we propose a compositional feature-based calibration method. Instead of directly using base prototypes our approach extracts and utilizes rich compositional information from the initial session to enhance new class representations. Experimental results on two keyword spotting datasets demonstrate the superiority of our proposed method showing improved performance in recognizing initial and new keywords.

Recently, neural vocoders have applied signal processing methods to synthesize speech to reduce computational complexity. However, most methods lack the benefits of a data-driven approach and the flexibility of hyper-parameters, such as filter length, because they rely on fixed signal processing filters. In this paper, we introduce Quad-Net, a network that includes restricted convolutional layers shaped by quadrature mirror synthesis filter banks. It is optimized with a perfect reconstruction loss derived from perfect reconstruction filter banks. This enables us to control filter lengths and degrees of data-drivenness. The results show that the filter parameters trained in our model exhibit characteristics similar to those of other signal processing methods with lower parameters. Furthermore, by increasing the filter length of Quad-Net, we can obtain filters that have complex frequency responses. It shows that a new approach enables the design of more complex filters that are adaptive to neural networks, diverging from previous methods.

Target sound extraction aims to isolate target sound sources from an input mixture using a target clue to identify the sounds of interest. To address the challenge posed by the wide variety of sounds, recent work has introduced privileged knowledge distillation (PKD), which utilizes privileged information (PI) about the target sound, available only during training. While PKD has shown promise, existing approaches often suffer from overfitting of the teacher model for the overly rich PI and ineffective knowledge transfer to the student model. In this paper, we propose Disentangled Codec Knowledge Distillation (DCKD) to mitigate these issues by regulating the amount and the flow of target sound information within the teacher model. We begin by extracting a compressed representation of the target sound using a neural audio codec to regulate the amount of PI. Disentangled representation learning is then applied to remove class information and extract fine-grained temporal information as PI. Subsequently, an n-hot vector as the class information and the class-independent PI are used to condition the early and later layers of the teacher model, respectively, forming a regulated coarse-to-fine target information flow. The resulting representation is transferred to the student model through feature-level knowledge distillation. Experimental results show that DCKD consistently improves existing methods across model architectures under the multi-target selection condition.

In deployment environments for speech recognition models, diverse proper nouns such as personal names, song titles, and application names are frequently uttered. These proper nouns are often sparsely distributed within the training dataset, leading to performance degradation and limiting the practical utility of the models. Personalization strategies that leverage userspecific information, such as contact lists or search histories, have proven effective in mitigating performance degradation caused by rare words. In this study, we propose a novel personalization method for combining the scores of a general language model (LM) and a personal LM within a probabilistic framework. The proposed method entails low computational costs, storage requirements, and latency. Through experiments using a realworld dataset collected from the vehicle environment, we demonstrate that the proposed method effectively overcomes the out-ofvocabulary problem and improves recognition performance for rare words

Zero-shot text-to-speech (TTS) has recently achieved remarkable performance by leveraging a speech prompt instead of a speaker embedding, as it provides richer information. However, zero-shot cross-lingual tasks synthesize speech in multiple languages according to a given language ID, regardless of the language of the speech prompt. Consequently, the inherent language-specific characteristics of the speech prompt may conflict with the language ID, potentially affecting the accuracy of language representation in speech. Thus, we propose vector field decomposition-based flow matching that decomposes the vector field into speaker and language components. These components are trained to be activated in different frequency bins, as speaker and language identity are distributed across distinct frequency ranges in speech. This approach is particularly effective for cross-lingual TTS, as it minimizes conflicts between speech prompts and language IDs. As a result, the summation of the two components directly forms the vector field that represents the probability path from a Gaussian distribution to the target data distribution (e.g., mel spectrogram). Experimental results demonstrate that the proposed method outperforms the conventional method in terms of both subjective and objective evaluations.

Recent advancements in packet loss concealment (PLC) have introduced diffusion-based generative models that offer high-quality audio reconstruction. However, their high computational costs make them impractical for real-time applications. In this letter, we present Flow-PLC, an efficient PLC model based on the flow-matching framework, designed to address these computational challenges. Flow-PLC achieves a remarkable 23× reduction in inference time compared to diffusion-based PLC models, requiring only five sampling steps to achieve near-optimal reconstruction. By significantly reducing computational complexity while maintaining high-quality results, Flow-PLC represents a substantial advancement in the development of efficient and practical generative PLC systems.

We propose a novel generative speech enhancement (SE) framework that integrates a language model (LM) and a flow-matching model. To utilize an LM with discrete tokens, we introduce dMel, which discretizes Mel spectrograms into a predefined set of quantized values on a linear-scale without requiring additional neural networks. dMel preserves both semantic and acoustic characteristics, providing a compact and effective token-based alternative to Mel spectrograms. We design the first encoder-decoder LM for SE, which learns to map noisy dMel to enhanced ones. Subsequently, flow-matching de-quantizes enhanced dMel into continuous representation and refines it by learning the optimal transport-based probability path, improving perceptual quality. This unified approach enables structured reconstruction while effectively suppressing noise. Experimental results demonstrate the effectiveness of our method in enhancing speech quality, establishing a new paradigm for generative SE without reliance on neural codec-based representations.

In this paper, we present the multi-rectangle inverse masking (MRIM), an extension and generalization of the traditional SpecAugment technique, for acoustic scene classification. While SpecAugment, observed from its unmasked areas, primarily forms rectangles around the input corners, our novel strategy generates rectangles at random positions with varied sizes, enhancing the data augmentation capacity. Our evaluations, conducted on the DCASE 2019 and 2020 datasets using CNN architectures like ResNet50 and BC-Res2Net, highlighted notable performance enhancements. Importantly, our method demonstrated resilience even when post-processing masking is applied to unseen test data, emphasizing its robustness across diverse acoustic scenes. To gain a deeper understanding of our method’s impact, we utilize the grad-CAM++ technique, a tool from explainable AI, to explore how masking influences model activations.

Score-based generative models have shown the real-like quality of synthesized speech in the text-to-speech (TTS) area. However, the critical artifact of score-based models is the requirement of a high computational cost due to the iterative sampling algorithm, and it also makes it difficult to fine-tune the score-based TTS-optimized vocoder. In this study, we propose a method of joint training the score-based TTS model and HiFi-GAN using the compressed log-mel features, and it guarantees a significant speech quality even on the non-iterative sampling. As a result, the proposed method overcomes some digital artifacts of the synthesized audios compared to the non-iterative sampling of Grad-TTS. Also, the non-iterative sampling can generate speech faster than other end-to-end TTS models with fewer parameters.

In this paper, we propose a timestamp knowledge distillation (TKD) method that adopts privileged knowledge distillation to enhance the performance of deep neural network (DNN)-based target sound extraction (TSE). While previous studies have mainly used n-hot vectors to indicate the type of target sound events (SEs), which are termed weak labels (WLs), recent studies demonstrated that timestamp knowledge of SEs is meaningful information to improve the TSE performance. To utilize timestamp knowledge, we use the oracle strong labels (OSLs) that indicate the occurrence of target SEs in the audio clip as privileged information. However, the OSLs are difficult to gain in real-world applications compared to WLs. We thus propose the TKD that transfers the timestamp knowledge from the teacher model trained using both WLs and OSLs to the student model trained using only WLs via a loss function. Experimental results across multiple DNN architectures confirmed that the OSLs enhanced the TSE significantly. Moreover, the TKD notably improved the student model’s performance compared to the baseline trained only with WLs.

Supervised deep learning has significantly improved bandwidth extension (BWE), whereas the emergence of self-supervised learning (SSL) has prompted the combined exploration of SSL and BWE. Although SSL-based deep learning models have shown to produce better representations than their supervised counterparts when trained naively, their effectiveness diminishes in when the model learns different tasks sequentially. To address this problem, we propose a continual learning framework called CLASS, which incorporates continual learning (CL) and self-supervised pretraining (SSP) to improve BWE performance. The framework integrates SSP and BWE fine-tuning tasks with CL approaches, enabling the model to retain its representation knowledge while adapting to BWE as a target task. We employ the CL fine-tuning loss or exponential moving average algorithm to gradually update model parameters and learn to resemble wideband from narrowband signals without losing information from a previous task. In addition, we present the new continual loss with extended version of elastic weight consolidation by updating fisher information matrix for better BWE performance. Our experimental results demonstrate that the proposed method outperforms the baseline approach on the TIMIT dataset. Furthermore, we explore the impact of different hyperparameter settings, contributing to a more comprehensive understanding of the performance of the proposed framework.

Research on personalizing neural transducer-based automatic speech recognition (ASR) systems using the text-only data is currently flourishing. Among various approaches, utilizing synthesized speech offers an advantage of adapting the entire ASR system. In this study, we explore the problem of personalization from a domain adaptation perspective and highlight the potential risk of overfitting associated with synthesized speech. To mitigate this risk, we propose the text-only unsupervised domain adaptation (ToUDA) strategy that robustly finetunes the generic ASR model on synthesized speech by incorporating parameter-averaging over time, model freezing, and filtering out-of-distribution instances. Via various experiments, we not only showcase the effectiveness of our approach but also uncover a noteworthy limitation when it comes to personalizing atypical speech.

Deep neural networks (DNNs) have revolutionized tasks such as image classification and speech recognition but often falter when training and test data diverge in distribution. External factors, from weather effects on images to varied speech environments, can cause this discrepancy, compromising DNN performance. Online test-time adaptation (OTTA) methods present a promising solution, recalibrating models in real-time during the test stage without requiring historical data. However, the OTTA paradigm is imperfect, often falling prey to issues such as catastrophic forgetting due to its reliance on noisy, self-trained predictions. Although some contemporary strategies mitigate this by tying adaptations to the static source model, this restricts model flexibility. This paper introduces a continual momentum filtering (CMF) framework, leveraging the Kalman filter (KF) to strike a balance between model adaptability and information retention. The CMF intertwines optimization via stochastic gradient descent with a KF-based inference process. This methodology not only aids in averting catastrophic forgetting but also provides high adaptability to shifting data distributions. We validate our framework on various OTTA scenarios and real-world situations regarding covariate and label shifts, and the CMF consistently shows superior performance compared to state-of-the-art methods.

In the rapidly evolving field of online test-time adaptation (OTTA), effectively managing distribution shifts is a pivotal concern. State-of-the-art OTTA methodologies often face limitations such as an inadequate target domain information integration, leading to significant issues like catastrophic forgetting and a lack of adaptability in dynamically changing environments. In this paper, we introduce a stationary latent weight inference (SLWI) framework, a novel approach to overcome these challenges. The proposed SLWI uniquely incorporates Bayesian filtering to continually track and update the target model weights along with the source model weight in online settings, thereby ensuring that the adapted model remains responsive to ongoing changes in the target domain. The proposed framework has the peculiar property to identify and backtrack nonlinear weights that exhibit local non-stationarity, thereby mitigating error propagation, a common pitfall of previous approaches. By integrating and refining information from both source and target domains, SLWI presents a robust solution to the persistent issue of domain adaptation in OTTA, significantly improving existing methodologies. The efficacy of SLWI is demonstrated through various experimental setups, showcasing its superior performance in diverse distribution shift scenarios.

Online test-time adaptation (OTTA) methods have demonstrated their effectiveness in real-time adapting to the target domain for speech recognition tasks.However, a common thread among these existing methods is their reliance on repetitive learning for each test utterance through a subloop, imposing prohibitive computational costs.This paper highlights the inefficiency inherent in applying a uniform number of subloop iterations to every test sample.To address this issue, we propose the online subloop search (OSS) method, which implicitly adjusts the number of iterations based on the test sample and domain characteristics.The proposed method operates within a framework comprising a chaser model updated via stochastic gradient descent and a leader model updated through the exponential moving average.The OSS method quantifies and quantizes the uncertainty in the chaser model relative to the leader model, using the quantized value to predict the number of iterations for the subloop.

We introduce diffusion-based packet loss concealment (DiffPLC), a novel approach designed to improve speech quality in the presence of packet losses for speech transmission. Derived from the foundation of a diffusion-based neural vocoder, the Diff-PLC introduces a crucial modification and supplementary concepts for the reconstruction of lost packets. A key aspect of the Diff-PLC involves integrating a feature-wise linear modulation layer into the diffusion model, facilitating the seamless incorporation of a conditioning feature. Furthermore, the Diff-PLC leverages packet loss embedding as an additional conditioning feature which significantly assists the diffusion model in restoring lost packets. The proposed model is evaluated using the blind test set of the INTERSPEECH 2022 PLC challenge, demonstrating the considerable restoration capabilities of Diff-PLC across various reference-free and reference-based metrics, including PLCMOS, PESQ, STOI, and NISQA.

In the rapidly evolving field of automatic speech recognition (ASR), the push towards personalization has become a paramount concern. Text-only personalization, while advantageous for data collection and adaptable to text variations, can suffer from overfitting when using personal data and requires extensive data to mitigate this issue. Federated learning (FL) emerges as a solution, facilitating learning from diverse client models while preserving privacy. However, FL addresses the challenges posed by non independent and identically distributed (non-i.i.d) data, potentially leading to poor performance. We propose two approaches for language model personalization in ASR to address these issues. First, adaptive weighted average addresses the limitations of uniform weight average in the existing FL method by combining local language models into a global model. Second, clustered federated learning, based solely on model parameters, improves model stability without relying on information from the local domain. Both strategies aim to enhance personalization and reduce performance degradation, particularly in non-i.i.d scenarios within the FL.

Most non-autoregressive text-to-speech (TTS) models acquire target phoneme duration (target duration) from internal or external aligners. They transform the speech-phoneme alignment produced by the aligner into the target duration. Since this transformation is not differentiable, the gradient of the loss function that maximizes the TTS model's likelihood of speech (e.g., mel spectrogram or waveform) cannot be propagated to the target duration. In other words, the target duration is produced regardless of the TTS model's likelihood of speech. Hence, we introduce a differentiable duration refinement that produces a learnable target duration for maximizing the likelihood of speech. The proposed method uses an internal division to locate the phoneme boundary, which is determined to improve the performance of the TTS model. Additionally, we propose a duration distribution loss to enhance the performance of the duration predictor. Our baseline model is JETS, a representative end-to-end TTS model, and we apply the proposed methods to the baseline model. Experimental results show that the proposed method outperforms the baseline model in terms of subjective naturalness and character error rate.

This letter presents a novel approach for accurately localizing moving object based on a robust time-of-arrival-based splitting mean positioning algorithm. The estimation performance of the existing localization method using the variational Bayesian Gaussian mixture model is degraded when a single observation is used. To overcome this limitation, the splitting mean online expectation maximization and closed-form solution are developed in this letter. The fundamental concept behind the proposed method involves splitting the mean of the approximate likelihood function into the true distance and bias components. These two components are estimated separately, enhancing the accuracy of localization. The simulation results demonstrate that the proposed algorithms outperform existing state-of-the-art methods in terms of localization performance.

Developing a lightweight speaker embedding extractor (SEE) is crucial for the practical implementation of automatic speaker verification (ASV) systems. To this end, we recently introducedbroadcasting convolutional neural networks (CNNs)-meet-vision-Transformers(BC-CMT), a lightweight SEE that utilizes broadcasted residual learning (BRL) within the hybrid CNN-Transformer architecture to maintain a small number of model parameters. We proposed three BC-CMT-based SEE with three different sizes: BC-CMT-Tiny, -Small, and -Base. In this study, we extend our previously proposed BC-CMT by introducing an improved model architecture and a training strategy based on knowledge distillation (KD) using self-attention (SA) maps. First, to reduce the computational costs and latency of the BC-CMT, the two-dimensional (2D) SA operations in the BC-CMT, which calculate the SA maps in the frequency–time dimensions, are simplified to 1D SA operations that consider only temporal importance. Moreover, to enhance the SA capability of the BC-CMT, the group convolution layers in the SA block are adjusted to have smaller number of groups and are combined with the BRL operations. Second, to improve the training effectiveness of the modified BC-CMT-Tiny, the SA maps of a pretrained large BC-CMT-Base are used for the KD to guide those of a smaller BC-CMT-Tiny. Because the attention map sizes of the modified BC-CMT models do not depend on the number of frequency bins or convolution channels, the proposed strategy enables KD between feature maps with different sizes. The experimental results demonstrate that the proposed BC-CMT-Tiny model having 271.44K model parameters achieved 36.8% and 9.3% reduction in floating point operations on 1s signals and equal error rate (EER) on VoxCeleb 1 testset, respectively, compared to the conventional BC-CMT-Tiny. The CPU and GPU running time of the proposed BC-CMT-Tiny ranges of 1 to 10 s signals were 29.07 to 146.32 ms and 36.01 to 206.43 ms, respectively. The proposed KD further reduced the EER by 15.5% with improved attention capability.

The performance of automatic speech recognition (ASR) models has been significantly improved owing to advances in deep learning and end-to-end approaches. However, these require a large amount of labeled data, which are expensive to obtain. Semi-supervised learning techniques, such as pseudo-labeling and self-supervised learning, have emerged as potential solutions to reduce the reliance on labeled data. Recently, some studies have combined self-supervised learning and pseudo-labeling to further enhance ASR performance. However, these methods suffer from incorrect pseudo-labels that propagate errors and reduce ASR performance. In this paper, we propose a novel method calledpartitioning attention weight(PAW) to mitigate the adverse effects of incorrect labels without requiring additional language models. Our proposed method isolates audio segments by partitioning a fully connected attention weight into sub-attention weights to prevent adverse effects that the model learns the wrong context for the entire attention weights from incorrect labels as well as overfitting. The proposed method is simple, requiring few changes to existing learning frameworks, and leverages the alignment information obtained during the pseudo-labeling process. Our experimental results show consistent performance improvements in ASR performance across various semi-supervised learning scenarios.

An attention-based automatic speech recognition (ASR) model generates a probability distribution of the tokens set at each time step. Recent studies have shown that calibration errors exist in the output probability distributions of attention-based ASR models trained to minimize the negative log likelihood. This study analyzes the causes of calibration errors in ASR model outputs and their impact on model performance. Based on the analysis, we argue that conventional methods for estimating calibration errors at the token level are unsuitable for ASR tasks. Accordingly, we propose a new calibration measure that estimates the calibration error at the sequence level. Moreover, we present a new post-hoc calibration function and training objective to mitigate the calibration error of the ASR model at the sequence level. Through experiments using the ASR benchmark, we show that the proposed methods effectively alleviate the calibration error of the ASR model and improve the generalization performance.

In this study, we extend self-distillation with no labels (DINO), a successful self-supervised learning framework, by combining it with supervised classification (SC) for semi-supervised speaker verification with limited labeled data. We introduce a transfer learning framework that pre-trains and fine-tunes the encoder using DINO and SC, respectively, and a multitask learning framework that shares the encoder while having separate projection layers for both methods. To achieve lower inter-speaker similarity, we propose a joint learning framework sharing both the encoder and projection layer for DINO and SC. We also propose an auxiliary contrastive loss between embeddings derived from labeled and unlabeled utterances and introduce a two-stage learning strategy to apply margin penalty effectively. Experimental results on the VoxCeleb corpus indicate that the joint learning framework outperforms the other frameworks and is closest to achieving the performance of fully supervised learning.

This paper addresses the critical challenge of distribution shifts in automatic speech recognition (ASR) systems through a novel framework, named adaptation without mode collapse (AWMC). Distribution shifts issue, where the source and target distributions differ, can severely degrade the performance of ASR systems. The proposed AWMC framework, a response to the issue, is designed to facilitate adaptive learning from sequentially streamed utterances and mitigates the effect of mode collapse, which is a common problem with traditional test-time adaptation (TTA) methodologies. Our framework employs three parameter-shared models (anchor, chaser, and leader) in concert to continually adapt from target data, significantly enhancing the performance and reliability of the system. The effectiveness of the AWMC is demonstrated through comprehensive performance comparisons with state-of-the-art TTA methods using widely recognized ASR datasets.

Due to the nature of neural networks that easily overfit the training set, neural network-based speech recognition models are vulnerable to prior shifts in data distribution or unseen words. Therefore, studies have been conducted to over-come this problem by using language models trained with a relatively easy-to-obtain unpaired corpus. In this paper, we present a new training method that uses BERT to improve the performance of a connectionist temporal classification (CTC)-based ASR model. The proposed method follows a cross-modal learning scenario and induces the CTC model to better embed contextual information by utilizing an auxiliary objective function operating at the sequence level. We applied the proposed method to fine-tune the pre-trained wav2vec 2.0 model with CTC loss and confirmed that the proposed method improves the generalization performance of the ASR model.

Sequence transducer is a popular end-to-end automatic speech recognition model for streaming scenarios: While, there is a trade-off between accuracy and latency. Latency regularization methods such as FastEmit can reduce latency, but the more they try to reduce latency, the worse accuracy tends to be. Conversely, knowledge distillation (KD) is only used to improve accuracy, and latency is not considered. In this paper, we propose an effective method that combines FastEmit with the KD to reduce latency and improve the accuracy of offline model in scenarios where the latency gap between offline and streaming models gets small. This method reduce the latency gap by applying with FastEmit to both the offline and streaming models. Experimental results on the LibriSpeech dataset show that the model with the best trade-off between accuracy and latency achieves a relative error reduction rate of 7.5% and reduces the latency by $130 \mathrm{~ms}$ compared with the streaming conformer transducer.

Despite the significant advancements and promising performance of deep learning-based packet loss concealment (PLC) systems in transmission systems, their focus on modeling acoustic features for reconstructing lost packets is insufficient to achieve smooth transitions during speech reconstruction. Therefore, to address this limitation, we propose integrating linguistic information derived from a speech recognition system as auxiliary features in the PLC system. By extracting ASR-guided representations and incorporating them using auxiliary loss, we successfully demonstrate a substantial improvement in the perceptual quality and intelligibility of the reconstructed speech. Our evaluation conducted on the wall street journal dataset further validates the effectiveness of our approach through experiments involving different packet loss rates and performance metrics.

Vehicle velocity prediction is an important task in the automotive industry because it can improve a car’s fuel economy and reduce emissions. Velocity prediction task has been studied for many years, and recently deep learning-based techniques have received a lot of attention. To accurately predict vehicle velocity, it is essential to analyze vehicle characteristics, driving patterns, and road conditions. Previously reported methods have not been able to consider driving patterns, which is the most crucial factor in predicting velocity. In this paper, we propose a model named "CAN2V," which effectively analyzes the vehicle characteristics and driving patterns in the encoder through multi-task learning. This model is an interpretable model of what input variables were used for velocity prediction through a variable selection network. Experimental results on a real driving dataset show that our proposed method outperforms the previous methods on a mean absolute error (MAE) and root-mean-square error (RMSE).

Representation models pre-trained on unlabeled data show competitive performance in speech recognition, even when fine-tuned on small amounts of labeled data. The continual representation learning (CTRL) framework combines pre-training and continual learning methods to obtain powerful representation. CTRL relies on two neural networks, online and offline models, where the fixed latter model transfers information to the former model with continual learning loss. In this paper, we present momentum continual representation learning (M-CTRL), a framework that slowly updates the offline model with an exponential moving average of the online model. Our framework aims to capture information from the offline model improved on past and new domains. To evaluate our framework, we continually pre-train wav2vec 2.0 with M-CTRL in the following order: Librispeech, Wall Street Journal, and TED-LIUM V3. Our experiments demonstrate that M-CTRL improves the performance in the new domain and reduces information loss in the past domain compared to CTRL.

Time-scale modification (TSM) is important in streaming services, including over-the-top (OTT) platforms, audiobooks, and online lectures. Although TSM modifies the speed of audio while maintaining other audio attributes such as the pitch and timbre of the speaker, it unnaturally distorts audio signals and makes spoken content difficult to understand. This study proposes an adaptive time-scale modification algorithm (ATSM); that adaptively varies the speaking rate for each phoneme cluster of speech to improve speech intelligibility. The proposed algorithm performs forced alignment using Montreal forced aligner and time-scale reconstruction using an adaptive speaking rate based on dynamic time warping. To validate the proposed algorithm, the diagnostic rhyme test (DRT) score, comparison mean opinion score (CMOS), and fast dynamic time warping (FastDTW) score of ATSM are compared with those of conventional TSMs. The results show that the speech compressed with the proposed algorithm has improved speech intelligibility than that of speech compressed with other algorithms.

Transformer-based end-to-end neural speaker diarization (EEND) models utilize the multi-head self-attention (SA) mechanism to enable accurate speaker label prediction in overlapped speech regions. In this study, to enhance the training effectiveness of SA-EEND models, we propose the use of auxiliary losses for the SA heads of the transformer layers. Specifically, we assume that the attention weight matrices of an SA layer are redundant if their patterns are similar to those of the identity matrix. We then explicitly constrain such matrices to exhibit specific speaker activity patterns relevant to voice activity detection or overlapped speech detection tasks. Consequently, we expect the proposed auxiliary losses to guide the transformer layers to exhibit more diverse patterns in the attention weights, thereby reducing the assumed redundancies in the SA heads. The effectiveness of the proposed method is demonstrated using the simulated and CALLHOME datasets for two-speaker diarization tasks, reducing the diarization error rate of the conventional SA-EEND model by 32.58% and 17.11%, respectively.

Representation-based speech recognition models have demonstrated state-of-the-art performance on downstream tasks. These models are pre-trained on large-scale unlabeled data, fine-tuned on a small amount of labeled data, and subsequently advanced via the self-training procedure by leveraging pseudo-labels. However, a self-trained representation model produces prediction errors caused by training with incorrect labels in the pseudo-labeled data. Weight-averaging methods have been employed to refine the pseudo-labels in a variety of studies; however, these methods amplify the prediction errors of each self-trained model. To alleviate this problem, we propose RepackagingAugment, a data augmentation method that improves the diversity of models while preventing the same incorrect labels from recursively occurring in every epoch. Our data augmentation deconstructs the paired speech–text data into word units and repackages them into a randomly determined number of word sequences. This strategy induces the models to produce different prediction errors by mitigating the problem of incorrect label over-fitting. Through various experiments on representation models, such as wav2vec 2.0 and data2vec, we demonstrate that our approach improves the performance of weight-averaged models.

Data augmentation using additive noise is a framework for robustly training automatic speech recognition models. To utilize noise information efficiently, previous studies used an additional branch to classify noise conditions. This added branch has a limited effect on the ASR because it performs independently of the ASR branch that classifies senones. In this paper, we propose a noise-aware target extension (NATE) that extends the senone target to contain noise awareness by jointly classifying the senone and noise in a single branch. In the inference stage, the output of the model is processed separately by the noise condition and then aggregated to match the senone posterior distribution. In addition, we combine NATE with self-distillation (NATEsd) to reduce the model parameters and avoid discrepancies between the outputs of training and inference. The effectiveness of the NATE method is validated on the two benchmark development and evaluation sets and simulated noisy test sets, resulting in significant improvements over the previous methods.

Enhancing speech quality and intelligibility for automatic speech recognition (ASR) plays an important role in modeling speech enhancement (SE) systems. However, improving the ASR performance by utilizing SE networks is not guaranteed, owing to the discrepancy in the training methods of the two systems. Therefore, recent studies have gradually incorporated ASR information into SE systems by jointly training ASR and SE systems. Although prior studies have improved the performance, they are inefficient because the two networks are combined and require large model sizes. To address this limitation, we propose an efficient way to use feature-wise linear modulation (FiLM) conditioning with CTC-based ASR probabilities for the SE system. The proposed model is designed by stacking a FiLM layer with selective learning on each temporal convolutional network of the SE estimation module. This allows the SE network to adaptively select ASR information based on the relationship between context and acoustic information. The proposed method improves SE and ASR performance, resulting in more robust results against noise with only a small increase in the number of parameters.

The remarkable performance of the pre-trained language model (LM) using self-supervised learning has led to a major paradigm shift in the study of natural language processing. In line with these changes, leveraging the performance of speech recognition systems with massive deep learning-based LMs is a major topic of speech recognition research. Among the various methods of applying LMs to speech recognition systems, in this paper, we focus on a cross-modal knowledge distillation method that transfers knowledge between two types of deep neural networks with different modalities. We propose an acoustic model structure with multiple auxiliary output layers for cross-modal distillation and demonstrate that the proposed method effectively compensates for the shortcomings of the existing label-interpolation-based distillation method. In addition, we extend the proposed method to a hierarchical distillation method using LMs trained in different units (senones, monophones, and subwords) and reveal the effectiveness of the hierarchical distillation method through an ablation study.

Spatial filtering of speech based on neural networks has been widely studied. However, existing approaches focus on improving signal extraction or separation performance, and how to define the signal in the direction-of-interest (DOI) for spatial filtering has not been investigated in detail. This study proposes a method to train neural networks for extracting directivity components of speech signals in the DOI. To this end, we formulate the problem by defining the DOI and its corresponding desired signal in a reverberant environment. Moreover, we demonstrate an on-the-fly training data generation procedure to feed the spatially diverse data to train the networks. The proposed method was evaluated with regard to spatial speech extraction and localization performance. In particular, it has been confirmed that the network trained with the proposed method using simulated datasets also functions for real recordings.

Deep learning-based speech synthesis evolves by employing a sequence-to-sequence (seq2seq) structure with an attention mechanism. The seq2seq speech synthesis model consists of a pair of the encoder for delivering the linguistic features and the decoder for predicting the mel-spectrogram, and learns the alignment between text and speech through the attention mechanism. The decoder predicts the mel-spectrogram by an autoregressive flow that considers the current input and what they have learned from previous inputs. This is beneficial when processing the sequential data, as in speech synthesis. However, the recursive generation of speech typically requires extensive training time, which slows the speed of synthesis. To overcome these obstacles, we propose a non-autoregressive framework for fully parallel deep convolutional neural speech synthesis. Firstly, we design a new synthesis paradigm that integrates a time-varying metatemplate (TVMT), whose length is modeled with a separate conditional distribution, to prepare the decoder input. The decoding step converts the TVMT into spectral features, which eliminates the autoregressive flow. Secondly, we propose a structure that uses multiple decoders interconnected by up-down chains with an iterative attention mechanism. The decoder chains distribute the burden of decoding, progressively infusing the information obtained from the training target example into the chains to refine the predicted spectral features at each decoding step. For each decoder, the attention mechanism is repeatedly applied to produce the elaborated alignment between the linguistic features and the TVMT, which is gradually transformed into the spectral features. The proposed architecture substantially improves the synthesis speed, and the resulting speech quality is superior to that of a conventional autoregressive model.

Speech dereverberation is an important issue for many real-world speech processing applications. Among the techniques developed, the weighted prediction error (WPE) algorithm has been widely adopted and advanced over the last decade, which blindly cancels out the late reverberation component from the reverberant mixture of microphone signals. In this study, we extend the neural-network-based virtual acoustic channel expansion (VACE) framework for the WPE-based speech dereverberation, a variant of the WPE that we recently proposed to enable the use of dual-channel WPE algorithm in a single-microphone speech dereverberation scenario. Based on the previous study, some ablation studies are conducted regarding the constituents of the VACE-WPE in an offline processing scenario. These studies reveal the characteristics of the system, thereby simplifying the architecture and leading to the introduction of new strategies for training the neural network for the VACE. Experimental results demonstrate that VACE-WPE (our PyTorch implementation and pre-trained models are available fromhttps://github.com/dreadbird06/vace_wpe) considerably outperforms its single-channel counterpart in simulated noisy reverberant environments in terms of objective speech quality and is superior to the single-channel WPE as well as several fully neural speech dereverberation methods when employed as the front-end for the far-field automatic speech recognizer.

Developing a single-microphone speech denoising or dereverberation front-end for robust automatic speaker verification (ASV) in noisy far-field speaking scenarios is challenging. To address this problem, we present a novel front-end design that involves a recently proposed extension of the weighted prediction error (WPE) speech dereverberation algorithm, the virtual acoustic channel expansion (VACE)-WPE. It is demonstrated experimentally in this study that unlike the conventional WPE algorithm, the VACE-WPE can be explicitly trained to cancel out both late reverberation and background noise. To build the front-end, the VACE-WPE is first (pre)trained to preserve the noise components in the input signals and produce “noisy” dereverberated output signals, thus making the front-end to be inductively biased to preserve as much noise components as possible and perform dereverberation only. Subsequently, given a pretrained speaker embedding model, the VACE-WPE is additionally fine-tuned within a task-specific optimization (TSO) framework, causing the speaker embedding extracted from the processed signal to be similar to that extracted from the “noise-free” target signal. Consequently, the front-end is optimized not to perform unnecessarily excessive denoising, thus achieving “generally safe” dereverberation and denoising for far-field ASV. Moreover, to prevent the front-end from adversely affecting the unconstrained “in-the-wild” ASV performance under more general, non-far-field conditions, we propose a distortion regularization method within the TSO framework. The effectiveness of the proposed approach is verified on both far-field and in-the-wild ASV benchmarks, demonstrating its superiority over fully neural front-ends and other TSO methods in various cases.

Short-utterance embedding, which is a speaker embedding extracted from a short utterance, shows poor speaker verification performance due to insufficient speaker information. To address the problem, we propose a method to map the set of short-utterance embeddings to a set of long-utterance embeddings based on a neural network. Specifically, a speech utterance is cropped into multiple segments whose durations are gradually increasing, and the speaker embeddings are extracted from the sequence of cropped segments using a pre-trained speaker embedding extractor. Subsequently, the sequence of embeddings is divided into a group of short-utterances embeddings and that of long-utterance embeddings. In our method, a sequence-to-sequence model based forecasting technique is employed, where an encoder transforms the group of short-utterance embeddings to a fixed-dimensional vector, and then a decoder converts the vector into a group of long-utterance embeddings. Experimental results on the VoxCeleb and Speakers in the Wild datasets show that our method improves the text-independent speaker verification performance under short utterance condition.

In this paper, a multi-input multi-output (MIMO) noise suppression (NS) algorithm for sound source localization (SSL) is proposed in mobile robot environment. Especially for the mobile robot, some ego noise (e.g., motor noise) is naturally generated and it becomes a dominant signal at microphone array, which seriously damages the performance of the SSL. Therefore, the proposed MIMO NS algorithm is designed not only to suppress the noise but also to preserve spatial information of sound sources. To this end, the recent MIMO time-domain audio separation network (TasNet) with preserving spatial cues for binaural source separation is extended to tetrahedral microphone array in the proposed MIMO NS algorithm. Furthermore, weighted phase error (WPE) between inter-channels is additionally employed in the loss function, which can improve the performance especially in preserving inter-channel time difference (ITD). The performance of the proposed approach is verified within the context of the robot vacuum cleaner, which shows from simulations and real experiments that the proposed approach can preserve the spatial information as well as remove the vacuum noise.

This study presents a robust two-step weighted least squares (WLS) localization algorithm using the MM estimator and the Kalman filter with the maximum Versoria criterion (MVC). An outlier-resistant statistic for the actual transformed distance is determined and the covariance matrix of the outlier-resistant statistic is calculated. This covariance matrix is used in the two-step WLS method. The simulation results demonstrate that the localization performances of the proposed algorithms outperform that of the conventional methods.

Herein, the authors present a robust estimator of range against the impulsive noise using only the received signal's magnitude. The M estimator has been widely used in robust signal processing. However, the existing M estimator requires statistical testing involving a threshold which has an optimality that varies with time, hence algorithmically challenging and computationally burdensome. The statistical testing is utilised for discerning the inlier and outlier. Further, statistical testing renders the computational burden of the algorithm high since the testing must be performed for each observation. Therefore, they propose the M estimator based on the hyper‐tangent loss function, which does not demand statistical testing. Conventional M estimator employing information theoretic learning also does not call for statistical testing, but the mean square error (MSE) performance for the range estimation is inferior to that of the proposed method. Furthermore, they perform an analysis for the MSE for the proposed algorithm. Monte Carlo simulations not only validate their theoretical analysis, but also demonstrate the MSE performance of the proposed method is nearly same as the existing skipped filter although it does not require the statistical testing and optimal threshold selection.

This paper presents robust localization algorithms that use range measurements to estimate the location parameters. The non-line-of-sight (NLOS) propagation of a signal can severely deteriorate the estimation performance in indoor and population-dense urban areas. Therefore, the robust localization algorithms are considered in this paper. In particular, the robust statistics-based localization is dealt with. The maximum likelihood (ML)-type and multi-stage ML-type method-based weighted least squares (WLS) algorithms and robust extrapolated single propagation unscented Kalman filter (ESPUKF) are proposed for mixed line-of-sight (LOS)/NLOS environments. Based on extensive simulations, the positioning accuracies of the proposed methods are found to be superior to those of conventional methods in the mildly and moderately mixed LOS/NLOS conditions. In addition, analyses are conducted on the mean square error (MSE), asymptotical unbiasedness and computational complexity of the proposed algorithms.

Nonlinear acoustic echo cancellation (AEC) is a highly challenging task in a single-microphone; hence, the AEC technique with a microphone array has also been considered to more effectively reduce the residual echo. However, these algorithms track only a linear acoustic path between the loudspeaker and the microphone array. This study proposes a microphone array form of the single-microphone nonlinear AEC (NAEC) algorithm in the reverberant condition. We extend a single-microphone-based model of the nonlinear acoustic echo to the microphone array case and propose the modeling of the acoustic transfer function (ATF) vector extended with a power series using a state-space equation. The Kalman filter is also adapted to optimally and recursively estimate the ATF vector. Furthermore, low-rank approximation and multi-microphone Wiener filtering are applied to estimate the multi-microphone near-end speech covariance, which results in the microphone array NAEC algorithm showing a consistently outstanding performance under severe signal-to-echo ratio (SER) and highly reverberant conditions. Consequently, our approach outperforms conventional methods regarding echo reduction and near-end speech quality for a wide range of SER and reverberation conditions.

The principal aim of this paper is to present a novel speech transmission system that conveys speech between an actuator in a wearable wristwatch and the ear bone of a user through a finger. If an individual wears a smart watch equipped with an actuator that can play speech sent via communication lines, speech vibrations propagate from the actuator to fingertips through the human tissue and bone. When an individual places his or her finger into their ear, speech conducted through the finger can be registered and heard. While listening to finger-conducted speech, sounds are muffled, significantly degrading the intelligibility of speech. To mitigate this problem, a formant enhancement filter is applied to the speech prior to being fed into the actuator. With this method, the impulse response of human tissue and bones between the fingertips and wrist on which the watch is worn is first estimated to account for speaker-dependent distortion. Based on the estimated impulse response, a gain filter is used to boost the sound spectra, especially within the formant regions, to compensate for frequency distortion prior to speech transmission. On the other hand, since the impulse responses of humans are quite different for each individual, we propose the novel idea of a personalized algorithm that guides users to select an appropriate gain filter, using the k-medoids clustering algorithm. Also, when an individual uses the proposed system, speech quality is degraded due to the ambient noise and acoustic echo between the microphone and actuator in the watch. Thus, to reduce background noise and acoustic echo, an integrated acoustic echo and background noise suppression algorithm is employed. Extensive simulations of the proposed system were performed by creating a novel phantom, which mimics the human hand with an aid of an ear simulator. We demonstrate that the proposed system has improved speech quality, when transmitting speech from the wearable wristwatch to a human perceptual organ through the finger.

Herein, we present robust shrinkage range estimation algorithms for which received signal strength measurements are used to estimate the distance between emitter and sensor. The concepts of robustness for the Hampel filter and skipped filter are combined with shrinkage for the positive blind minimax and Bayes shrinkage estimation. It is demonstrated that the estimation accuracies of the proposed methods are higher than those of the existing median-based shrinkage methods through extensive simulations.

In this paper, a deep neural network (DNN) based integrated background noise suppression and acoustic modeling for speech recognition proposed in which on/off state of the motor for the head shaking robot is employed as the relevant auxiliary information of the DNN input. Since the motor sound being generated when the robot is moving or shaking its head severely degrades the performance of the speech recognition accuracy, we propose to use the motor on/off state as additional information when designing the DNN-based recognition system. Our speech recognition algorithm consists of two parts including the feature mapping model for feature enhancement and the acoustic model for phoneme recognition. As for the feature mapping, the stacked DNN is designed for the precise feature enhancement such that the lower DNN and upper DNN are trained separately and combined after which the motor state is plugged into both the lower DNN and upper DNN in addition to the input noisy speech. Then, the acoustic model is trained upon the feature enhancement model in which the motor state is again used as the augmented feature. The proposed technique to suppress the acoustic and motor noises was evaluated in term of the phoneme error rate (PER) and showed a significant improvement over the conventional system.

This study presents the shrinkage‐based sequential source localisation and range estimation algorithms. The shrinkage factor is found using the variance of the estimate in the existing shrinkage algorithm. However, the variance of the estimate is difficult to calculate when the form of the estimate is complex. To circumvent this problem, the authors propose a shrinkage algorithm that employs the Cramér–Rao lower bound (CRLB) instead of the variance for the maximum likelihood (ML) estimate. The variance of the ML estimate and CRLB were found to be similar in simulation results. Furthermore, Stein's unbiased risk estimator and Ledoit–Wolf methods are used to determine the shrinkage factor. The resulting estimation accuracy of the proposed shrinkage‐based sequential source localisation and range estimation methods was similar with that of the existing shrinkage algorithm.

Oscillometric measurement is widely used to estimate systolic blood pressure (SBP) and diastolic blood pressure (DBP). In this paper, we propose a deep belief network (DBN)-deep neural network (DNN) to learn about the complex nonlinear relationship between the artificial feature vectors obtained from the oscillometric wave and the reference nurse blood pressures using the DBN-DNN-based-regression model. Our DBN-DNN is a powerful generative network for feature extraction and can address to stick in local minima through a special pretraining phase. Therefore, this model provides an alternative way for replacing a popular shallow model. Based on this, we apply the DBN-DNN-based regression model to estimate the SBP and DBP. However, there are a small amount of data samples, which is not enough to train the DBN-DNN without the overfitting problem. For this reason, we use the parametric bootstrap-based artificial features, which are used as training samples to efficiently learn the complex nonlinear functions between the feature vectors obtained and the reference nurse blood pressures. As far as we know, this is one of the first studies using the DBN-DNN-based regression model for BP estimation when a small training sample is available. Our DBN-DNN-based regression model provides a lower standard deviation of error, mean error, and mean absolute error for the SBP and DBP as compared with the conventional methods.

In this paper, we propose to improve the dual-microphone voice activity detection (VAD) technique for which a discriminative weight training is applied to achieve optimally weighted spatial features. In our approach, we first derive the maximum a posteriori (MAP) probabilities from the spatial features such as the power level difference ratio (PLDR), phase vector, and coherence. Then, we combine each MAP probability within the minimum classification error (MCE) framework to offer an optimal VAD decision in a spectral domain. Experimental results show that the proposed dual-microphone VAD algorithm shows better performances than the conventional dual-microphone VAD methods, which solely utilize the PLDR, phase, and spectral coherence.

In this study, the authors propose a closed‐form time‐of‐arrival source localisation method and justify the employment of the invariance property of the maximum likelihood (ML) estimator in the source localisation context with multiple samples. The magnitude of the bias of the proposed sample vector function (the statistic that consists of the multiple observations set) using the invariance property of the ML estimator is smaller than that based on the sample mean. Therefore, the mean squared error (MSE) of the weighted least squares estimate using the proposed sample vector function is smaller than that based on the sample mean when the variances of both sample vector functions are the same. Furthermore, the authors investigate a situation in which sensors have erroneous position information. The simulation results show that the averaged MSE performance of the proposed method is superior to that of the existing methods irrespective of the number of samples.

In this paper, we propose the regression-based packet loss concealment (PLC) for digital speech transmission by using deep neural networks (DNNs) with a multiple-layer deep architecture. For the DNN training, log-power spectra and phases are employed as features in the input layer for the large training set, which ensures non-linear mapping the frames from the last correctly received frame to the missing frame. Once the training is accomplished by the restricted Boltzmann machine (RBM)-based pre-training to initialize the DNN, minimum mean square error (MMSE)-based fine tuning is then performed based on the back-propagation algorithm. In the reconstruction stage, the trained DNN model is fed with the features of the previous frames in order to estimate the log-power spectra and phases of the missing frames. Reconstruction is further improved by using the cross-fading technique to mitigate discontinuity between the reconstruction signal and good frame signal in the time-domain. To demonstrate the performance of the proposed algorithm, hidden Markov model (HMM)-based PLC algorithm and the PLC algorithm standardized in adaptive multi-rate wideband (AMR-WB) Appendix I were used for comparison. The experimental results show that the proposed approach provides better speech quality and speech recognition accuracy than the conventional approaches.

Although the systolic and diastolic blood pressure ratios (SBPRs and DBPRs) based on the conventional maximum amplitude algorithm (MAA) are assumed to be fixed; this assumption is not valid. In this paper, we present an improved Gaussian mixture regression (IGMR) approach that can accurately measure blood pressure. The SBPR and DBPR are estimated by using the IGMR technique. Specifically, the number of feature's samples in the clustered feature space is increased using the nonparametric bootstrap technique to create the pseudo feature. The pseudo feature vector is much more matched than the original feature for the Gaussian mixture model (GMM) to fit individual BP characteristics in the training stage. By using the classified targeting clusters, we eventually estimate the SBPR and DBPR based on the IGMR technique at the test stage. The mean error (ME) and standard deviation of the error (SDE), and mean absolute error (MAE) of the SBP and DBP estimates obtained with the SBPR and DBPR using the proposed technique approaches are superior to the ME, SDE, and MAE of the estimates obtained using the conventional methods. The difference in the SDE between the proposed technique and the conventional MAA technique for the SBP and DBP turned out to be 3.67 and 3.08 mmHg in the simulation.

This paper presents two closed-form localization algorithms, a general algorithm and a colocated algorithm, for distributed multiple-input multiple-output (MIMO) radar systems. In distributed MIMO radar systems, range sum measurements are used to estimate the location parameter. For this, the range sum error minimization is actually employed to be degenerated into two cases for time-of-arrival (TOA) passive localization, one by employing the distance estimate between the target and the receiver (the general algorithm) and the other by subtracting the distance measurement between the target and the transmitter after the time-delay estimation (the colocated algorithm). The resulting positioning accuracy of the general and colocated techniques is found to perform better than that of the existing closed-form weighted least squares (WLS) algorithm and attain the Cramér-Rao lower bound (CRLB).

In diverse engineering problems including wireless communications, the estimate of the signal‐to‐noise ratio (SNR) is required. In this study, the authors develop a shrinkage‐based SNR estimator in the data‐aided and non‐data‐aided schemes for higher M ‐ary phase‐shift‐keying ( M ≥ 8) and quadrature amplitude modulations. The observed Cramér‐Rao lower bound is used as the variance of the expectation maximisation estimator to determine the optimal shrinkage factor. Simulation results show that the normalised mean‐squared error of the proposed method is lower than that of the expectation maximisation method for low and moderate SNR conditions.

In this letter, we propose a novel frequency-domain second-order Volterra filter based on soft decision for nonlinear acoustic echo suppression. This letter offers an efficient algorithm for nonlinear echo power estimation using the second-order Volterra filter and an AES algorithm using the estimated nonlinear echo power spectrum within a soft decision baseline by incorporating the ratio of the a priori probability of near-end speech presence and absence, which is obtained by using a data-driven training method.

In this paper, we propose a novel dual-microphone voice activity detection (VAD) technique based on the two-step power level difference (PLD) ratio. This technique basically exploits the PLD between the primary microphone and the secondary microphone in a mobile device when the distance between the microphones and the sound source is relatively short. Based on the PLD, we propose the use of the PLD ratio (PLDR) instead of the original PLD to take advantage of the relative difference between the PLD of speech and the PLD of noise. Indeed, the PLDR is obtained by estimating the ratio of the PLD between the input signals and the PLD between the two channel noises during periods without speech. The proposed technique offers a two-step algorithm using the PLDRs including long-term PLDR (LT-PLDR), which characterizes long-term evolution and short-term PLDR (ST-PLDR), which characterizes short-time variation during the first step. LT-PLDR-based and ST-PLDR-based VAD decision are performed using the maximum a posteriori (MAP) probability derived from the model-trust algorithm and combined at the second step to reach a superior VAD decision for both long-term and short-term situations. Extensive experimental results show that the proposed dual-microphone VAD technique outperforms the conventional two-channel VAD method as well as most standardized VAD algorithms.

In this paper, we propose an improved adaptive muting method using a sigmoid function for the packet loss concealment algorithm of ITU-T G.722 Recommendation. The packet loss concealment algorithm performs an adaptive muting to prevent the generation of unnecessary noise during packet loss recovery. While muting is linearly and discontinuously performed according to packet errors, our muting approach is performed by the non-linear and continuous sigmoid function. The principal parameters of the sigmoid function are obtained based on training at which minimization between the desired signal and the reconstructed signal is performed. Experimental results show that this proposed muting technique can enhance the performance of the packet loss concealment algorithm of G.722 under various packet loss environments.

In this paper, we propose a spectral difference approach for noise power estimation in speech enhancement. The noise power estimate is given by recursively averaging past spectral power values using a smoothing parameter based on the current observation. The smoothing parameter in time and frequency is adjusted by the spectral difference between consecutive frames that can efficiently characterize noise variation. Specifically, we propose an effective technique based on a sigmoid-type function in order to adaptively determine the smoothing parameter based on the spectral difference. Compared to a conventional method, the proposed noise estimate is computationally efficient and able to effectively follow noise changes under various noise conditions.

Variable bit-rate coding introduced for effective utilization of limited communication bandwidth requires accurate classification of input signals. This paper investigates implementation of a support vector machine (SVM)-based speech/music classifier in the selectable mode vocoder (SMV) framework, which is a standard codec adopted by the Third-Generation Partnership Project 2 (3GPP2). A support vector machine is well known for its superior pattern recognition capability; however, it is accompanied by a high computational cost. In order to achieve a more practical system, three techniques are proposed for the SVM-based speech/music classifier. The first is to prune support vectors that least contribute to the output of the SVM, while the other two are aimed at reducing the number of classification requests to the SVM-based classifier by eliminating or redirecting some of the classification requests to the classifier.

Support vector machines (SVMs) have been recognised as a promising technique in the field of pattern recognition, and one of the interesting applications of this technique is speech/music classification. In this study, the authors propose a novel approach to improve the SVM-based speech/music classification using the second-order conditional maximum a posteriori (CMAP). To do this, the authors first devise a method to estimate a posteriori probability to select between speech and music from the SVM output. This is achieved by employing the sigmoid function, obtained by optimised data training. A final speech/music classification is then acquired using the second-order CMAP with a maximum a posteriori probability depending not only on the current observation, but also on the classification results of two previous frames, incorporating substantial inter-frame correlations. While conventional SVM optimisation techniques are used during the training phase, the proposed technique can be inherently adopted in the classification phase. In this regard, the proposed approach can be developed and employed in parallel with other optimisation techniques. Experimental results show that the proposed algorithm yields better results than the speech/music classification rule in SVM.

In this letter, we present a speech enhancement technique based on the ambient noise classification incorporating the Gaussian mixture model (GMM). The principal parameters of the statistical model-based speech enhancement algorithm such as the weighting parameter in the decision-directed (DD) method and the long-term smoothing parameter of the noise estimation, are chosen as different values according to the classified contexts to ensure best performance for each noise. For the real-time environment awareness, the noise classification is performed on a frame-by-frame basis using the GMM with the soft decision framework. Thespeechabsenceprobability(SAP)isusedindetecting the speech absence periods and updating the likelihood of the GMM. Index Terms: Speech Enhancement, Noise Classification, Soft Decision, Gaussian Mixture Model

In this paper, we propose a novel double-talk detection (DTD) technique based on a soft decision in the frequency domain. The proposed method provides an efficient procedure to detect the double-talk situation by the use of the global near-end speech presence probability (GNSPP) and voice activity detection (VAD) of the near-end and far-end signal. Specifically, the GNSPP is derived based on a statistical method of speech and is employed to determine the double-talk presence in a given frame. The performance of our approach is evaluated by objective tests under different environments, and it is found that the suggested method yields better results compared with the conventional scheme.

In this paper, we propose a novel approach to statistical modelbased voice activity detection (VAD) that incorporates a secondorder conditional maximum a posteriori (MAP) criterion. As a technicalimprovementforthefirst-orderconditionalMAPcriterionin[1],weconsiderboththecurrentobservationandthevoice activity decision in the previous two frames to take full consideration of the inter-frame correlation of voice activity. The soft decision scheme is incorporated to result in time-varying thresholdsforfurtherperformanceimprovement. Experimentalresults show that the proposed algorithm outperforms the conventional CMAP-based VAD technique under various experimental conditions.

In this paper, we propose a novel frequency-domain approach to double-talk detection (DTD) based on the Gaussian mixture model (GMM). In contrast to a previous approach based on a simple and heuristic decision rule utilizing time-domain crosscorrelations, GMM is applied to a set of feature vectors extracted from the frequency-domain cross-correlation coefficients. Performance of the proposed approach is evaluated through objective tests under various environments, and better results are obtained as compared to the time-domain method. Index Terms: Voice Activity Detection, Second-order Conditional MAP, Soft Decision, Likelihood Ratio Test

In this letter, we propose a novel frequency-domain approach to double-talk detection (DTD) based on the Gaussian mixture model (GMM). In contrast to a previous approach based on a simple and heuristic decision rule utilizing time-domain cross-correlations, GMM is applied to a set of feature vectors extracted from the frequency-domain cross-correlation coefficients. Performance of the proposed approach is evaluated through objective tests under various environments, and better results are obtained as compared to the time-domain method.

In this paper, we propose a novel approach to improve the performance of minima controlled recursive averaging (MCRA) based on a conditional maximum a posteriori (MAP) criterion. From an investigation of the MCRA scheme, it is discovered that MCRA method cannot take full consideration of the inter-frame correlation of voice activity since the noise power estimate is adjusted by the speech presence probability depending on an observation of the current frame. To avoid this phenomenon, the proposed MCRA approach incorporates the conditional MAP criterion in which the noise power estimate is obtained using the speech presence probability conditioned on both the current observation and the speech activity decision in the previous frame Experimental results show that the proposed MCRA technique based on conditional MAP yields better results compared to the conventional MCRA method.

In this paper, we propose a novel approach to noise power estimation for robust noise suppression in noisy environments. From investigation of the state-of-the-art techniques for noise power estimation, it is discovered that the previously known methods are accurate mostly either during speech absence or speech presence but none of it works well in both situations. Our approach combines minimum statistics (MS) and soft decision (SD) techniques based on probability of speech absence. The performance of the proposed approach is evaluated by a quantitative comparison method and subjective test under various noise environments and found to yield better results compared with conventional MS and SD-based schemes. Index Terms: Minimum Statistics, Soft Decision

In this paper, we propose a novel acoustic echo suppression (AES) technique based on soft decision in a frequency domain. The proposed approach provides an efcient and unied frame- work for such procedures as AES gain computation, AES gain modication using soft decision, and estimation of relevant pa- rameters based on the same statistical model assumption of the near-end and far-end signal instead of the conventional strate- gies requiring the additional residual echo suppression (RES) step. Performances of the proposed AES algorithm are evalu- ated by objective tests under various environments and better results compared with the conventional AES method are ob- tained.

In this letter, we propose a novel speech enhancement technique based on global soft decision incorporating a support vector machine (SVM). Global soft decision in the proposed approach is performed employing the probabilistic outputs of the SVM rather than the conventional Bayes' rule. Actually, global speech absence probability (GSAP) is determined by the sigmoid function based on key parameters estimated by the model-trust minimization algorithm of the SVM output. Improved results are obtained in terms of speech quality measures for various types of noise and at different signal-to-noise ratio (SNR) levels when the proposed SVM is adopted in the global soft decision for speech enhancement.

In this letter, we propose a novel method to improve the minima controlled recursive averaging (MCRA) based on the second-order conditional maximuma posteriori(CMAP). From an investigation of the previous MCRA scheme, it is discovered that the MCRA method cannot take full consideration of the inter-frame correlation of voice activity since the noise power estimate is adjusted by the speech presence probability depending on a current observation. To avoid this phenomenon, we propose the MCRA approach incorporating the second-order CMAP criterion in which the noise power estimate is obtained using the speech presence probability conditioned on both the current observation and the speech activity decision in the previous two frames. Experimental results show that the proposed MCRA technique based on second-order CMAP yields better results compared to the previous MCRA method in speech enhancement.

In this letter, we propose a novel acoustic echo suppression (AES) technique based on soft decision in a frequency domain. The proposed approach provides an efficient and unified framework for such procedures as AES gain computation, AES gain modification using soft decision, and estimation of relevant parameters based on the same statistical model assumption of the near-end and far-end signal instead of the conventional strategies requiring the additional residual echo suppression (RES) step. Performances of the proposed AES algorithm are evaluated by objective and subjective tests under various environments, and better results compared with the conventional AES method are obtained.

In this paper, we apply a discriminative weight training to a statistical model-based voice activity detection (VAD). In our approach, the VAD decision rule is expressed as the geometric mean of optimally weighted likelihood ratios (LRs) based on a minimum classification error (MCE) method. That approach is different from that of previous works in that different weights are assigned to each frequency bin and is considered to be more realistic. According to the experimental results, the proposed approach is found to be effective for the statistical model-based VAD using the LR test.

One of the key issues in practical speech recognition is to achieve robust gender identification. Most conventional gender identification approaches use relevant features derived from the magnitude spectrum. In this paper, we propose a novel gender identification method using a group delay function (GDF). Based on the statistical analysis of the GDF, it is found that the GDF is an effective feature for gender identification. The experimental results demonstrate that the proposed method gives significant improvement compared to conventional methods.

In this letter, we apply a discriminative weight training to a statistical model-based voice activity detection (VAD). In our approach, the VAD decision rule is expressed as the geometric mean of optimally weighted likelihood ratios (LRs) based on a minimum classification error (MCE) method. That approach is different from that of previous works in that different weights are assigned to each frequency bin and is considered to be more realistic. According to the experimental results, the proposed approach is found to be effective for the statistical model-based VAD using the LR test.

In this letter, we propose a novel approach to noise power estimation for robust speech enhancement in noisy environments. From investigation of the state-of-the-art techniques for noise power estimation, it is discovered that the previously known methods are accurate mostly either during speech absence or speech presence, but none of it works well in both situations. Our approach combines minimum statistics (MS) and soft decision (SD) techniques based on probability of speech absence. The performance of the proposed approach is evaluated by a quantitative comparison method and subjective test under various noise environments and found to yield better results compared with conventional MS- and SD-based schemes.

In this letter, a novel approach is proposed to improve the performance of speech/music classification for the selectable mode vocoder (SMV) of 3GPP2 using the Gaussian mixture model (GMM). An in-depth analysis of the features and classification method adopted in the conventional SMV is performed. Feature vectors applied to the GMM are then selected from the relevant parameters of the SMV for efficient speech/music classification. The performance of the proposed algorithm is evaluated under various conditions and yields better results compared with the conventional scheme implemented in the SMV.

In this paper, we propose effective feature vectors to improve the performance of voice activity detection (VAD) employing a support vector machine (SVM), which is known to incorporate an optimized nonlinear decision over two different classes. To extract the effective feature vectors, we present a novel scheme combining the a posteriori SNR, a priori SNR, and predicted SNR, widely adopted in conventional statistical model-based VAD. Based on the results of experiments, the performance of the SVM-based VAD using novel feature vectors is found to be better than that of ITU-T G.729B and other recently reported methods.

This letter presents a new voice activity detection (VAD) method using a uniformly most powerful (UMP) test to improve the performance of Sohn's et al.'s VAD. The UMP test is employed to derive a redefined decision rule based on the likelihood ratio test (LRT). The proposed method provides the decision rule by comparing the magnitude of the noisy speech signal to the adaptive threshold estimated from the noise statistics. Experimental results show that the proposed VAD algorithm based on the UMP test outperforms those based on the widely used Gaussian distribution.

Abstract In this paper, an improved preprocessor for low-bit-rate speech coding employing the perceptual weightingfilter is proposed. Speech modification in the proposedapproach is performed according to a criterion whichmakes a compromise between the modification and per-ceptual weighted quantization errors. For this, the per-ceptual weighting filter is expressed in terms of a trans-form domain matrix. The proposed approach is effec-tive in enhancing the speech signal at coder-decoder(CODEC) output through a number of listening tests. 1. Introduction Ingeneral,theperformanceofalow-bit-ratespeechcoderdegrades seriously under the presence of various inter-fering signals such as background noise, acoustic echo,music sounds or interfering speaker’s speech. This phe-nomenon is mainly due to the deviation from the as-sumed speech production model which is used in thecodebook training since a number of codebooks used inthe coder are trained based on a large amount of speechdata and the ranges for parameter search are specified tofit the pure speech signals. One of the successful appli-cations of the unwanted distortion reduction technique tolow-bit-rate coding is the speech enhancement technique[2, 3, 8, 1]. Even though aforementioned enhancementtechniques have been found effective in the presence of astationary background noise, they are not capable of han-dling such interfering signals as the acoustic echoes, mu-sicsoundsorco-talkers’speech. Thisismainlyduetothefactthattheconventionalapproachesadopttheopenloopanalysis which can not take advantage of speech codercharacteristics. An alternative method is the generalized

In this paper, we apply a new structural approach to generalized analysis-by-synthesis (GAbS) for system identification as a preprocessor of a low-bit-rate speech coder. In our approach, the coder-decoder (CODEC) system is separately estimated and then applied to modify the current input signal. This is different from that originally proposed where the CODEC system is sequentially estimated and then applied to the next input signal. The proposed estimation scheme is compared to the conventional method in terms of the signal modification approach under the various noise data and in several SNR conditions, and shows better performance.

We propose a voice activity detection (VAD) algorithm based on the generalized gamma distribution (G/spl Gamma/D). The distributions of noise spectra and noisy speech spectra, including speech-inactive intervals, are modeled by a set of G/spl Gamma/Ds and applied to the likelihood ratio test (LRT) for VAD. The parameters of G/spl Gamma/D are estimated through an on-line maximum likelihood (ML) estimation procedure where the global speech absence probability (GSAP) is incorporated under a forgetting scheme. Experimental results show that the proposed VAD algorithm, based on G/spl Gamma/D, outperformed the algorithms based on other statistical models.

In this paper, we apply a new structural approach to generalized analysis-by-synthesis (GAbS) for system identification as a preprocessor of a low-bit-rate speech coder. In our approach, the coder-decoder (CODEC) system is separately estimated and then applied to modify the current input signal. This is different from that originally proposed where the CODEC system is sequentially estimated and then applied to the next input signal. The proposed estimation scheme is compared to the conventional method in terms of the signal modification approach under the various noise data and in several SNR conditions, and shows better performance.

In this letter, we propose results of distribution tests that indicate that for many natural images, the statistics of the discrete cosine transform (DCT) coefficients are best approximated by a generalized gamma function (G/spl Gamma/F), which includes the conventional Gaussian, Laplacian, and gamma probability density functions. The major parameter of the G/spl Gamma/F is estimated according to the maximum likelihood (ML) principle. Experimental results on a number of /spl chi//sup 2/ tests indicate that the G/spl Gamma/F can be used effectively for modeling the DCT coefficients compared to the conventional Laplacian and generalized Gaussian function (GGF).

In this letter, we propose a new statistical model, two-sided generalized gamma distribution (G/spl Gamma/D) for an efficient parametric characterization of speech spectra. G/spl Gamma/D forms a generalized class of parametric distributions, including the Gaussian, Laplacian, and Gamma probability density functions (pdfs) as special cases. We also propose a computationally inexpensive online maximum likelihood (ML) parameter estimation algorithm for G/spl Gamma/D. Likelihoods, coefficients of variation (CVs), and Kolmogorov-Smirnov (KS) tests show that G/spl Gamma/D can model the distribution of the real speech signal more accurately than the conventional Gaussian, Laplacian, Gamma, or generalized Gaussian distribution (GGD).

Usually, the performance of a low-bit-rate speech coder degrades seriously in the presence of various interfering signals such as the background noise, acoustic echo, co-talkers' speech and other unwanted signals. This comes from the mismatch between the input signal and the assumed speech production model on which the design of the given speech coder is based. In this paper, we present an approach to modify the input signal such that it can be coded more effectively within the generalized analysis-by-synthesis framework. Signal modification in the presented approach is performed according to a criterion which makes a compromise between the modification and coder quantization errors. The coder-decoder (CODEC) characteristic is described in terms of a transfer matrix, and an on-line method using the recursive least square (RLS) technique is proposed to estimate it. Since each part of the speech signal is differently affected by the modification, we also devise an adaptive method based on the signal-to-quantization noise ratio (SQNR). In contrast to the conventional modification techniques, our approach can be implemented as a simple front-end for any analysis-by-synthesis type coders.

This paper proposes a voice activity detector (VAD) based on the complex Laplacian model. With the use of a goodness-of-fit (GOF) test, it is discovered that the Laplacian model is more suitable to describe noisy speech distribution than the conventional Gaussian model. The likelihood ratio (LR) based on the Laplacian model is computed and then applied to the VAD operation. According to the experimental results, we can find that the Laplacian statistical model is more suitable for the VAD algorithm compared to the Gaussian model.

In this paper, we propose an approach to modify the input signal such that it can be coded more effectively within the generalized analysis-by-synthesis framework. Since most of the low-bit-rate speech coders are designed based on the human speech production mechanism, the perceived quality of the speech reconstructed in the decoder degrades seriously if the original input signal deviates from the pure speech. In order to alleviate this problem, we introduce a criterion which compromises the quantization error with the distortion incurred due to the modification. The coder-decoder (CODEC) characteristic is described in terms of a transfer matrix, and it is estimated according to the least squares criterion. In contrast to the conventional modification techniques, our approach can be implemented as a simple front-end for any analysis-by-synthesis type coders. The proposed approach is found effective in reducing audible distortions through a number of listening tests.

In this letter, we propose a preprocessor that modifies the input signal such that it can be coded more effectively in a low-bit-rate speech coder. Since most of the low-bit-rate speech coders are designed based on the human speech production mechanism, the perceived quality of the speech reconstructed in the decoder degrades seriously if the original input signal deviates from the pure speech. In order to alleviate this problem, we introduce a criterion that compromises the quantization error with the distortion incurred due to the modification. The coder-decoder characteristic is described in terms of a transfer matrix, and it is estimated according to the least squares criterion. The proposed approach is found to be effective in reducing audible distortions through a number of listening tests.

In this letter, we propose a novel speech enhancement technique based on global soft decision. The proposed approach provides a unified framework for such procedures as speech absence probability (SAP) computation, spectral gain modification, and noise spectrum estimation using the same statistical model assumption. Performances of the proposed enhancement algorithm are evaluated by subjective tests under various environments and show better results compared with the IS-127 standard enhancement method.