It has been shown that the intelligibility of noisy speech can be improved by speech enhancement (SE) algorithms. However, monaural SE has not been established as an effective frontend for automatic speech recognition (ASR) in noisy conditions compared to an ASR model trained on noisy speech directly. The divide between SE and ASR impedes the progress of robust ASR systems, especially as SE has made major advances in recent years. This paper focuses on eliminating this divide with an ARN (attentive recurrent network) time-domain, a TF-CrossNet time-frequency domain, and an MP-SENet magnitude-phase based enhancement model. The proposed systems decouple frontend enhancement and backend ASR, with the latter trained only on clean speech. Results on the WSJ, CHiME-2, LibriSpeech, and CHiME-4 corpora demonstrate that ARN, TF-CrossNet, and MP-SENet enhanced speech all translate to improved ASR results in noisy and reverberant environments, and generalize well to real acoustic scenarios. The proposed system outperforms the baselines trained on corrupted speech directly. Furthermore, it cuts the previous best word error rate (WER) on CHiME-2 by 28.4% relatively with a 5.6% WER, and achieves 3.3/4.4% WER on single-channel CHiME-4 simulated/real test data without training on CHiME-4. We also observe consistent improvements using noise-robust Whisper as the backend ASR model.

Deploying speech enhancement (SE) systems in wearable devices, such as smart glasses, is challenging due to the limited computational resources on the device. Although deep learning methods have achieved high-quality results, their computational cost limits their feasibility on embedded platforms. This work presents an efficient end-to-end SE framework that leverages a Differentiable Digital Signal Processing (DDSP) vocoder for high-quality speech synthesis. First, a compact neural network predicts enhanced acoustic features from noisy speech: spectral envelope, fundamental frequency (F0), and periodicity. These features are fed into the DDSP vocoder to synthesize the enhanced waveform. The system is trained end-to-end with STFT and adversarial losses, enabling direct optimization at the feature and waveform levels. Experimental results show that our method improves intelligibility and quality by 4% (STOI) and 19% (DNSMOS) over strong baselines without significantly increasing computation, making it well-suited for real-time applications.

This paper investigates robust speaker localization at the frame level on the basis of complex spectral mapping, which is capable of learning both the magnitude and phase of the target signal. Unlike prevailing deep learning methods for speaker localization, we perform MIMO (multi-input multi-output) based multi-channel speech enhancement first and then localize the enhanced speaker using weighted generalized cross correlation. In addition, we propose new multi-channel loss functions that incorporate phase differences in order to preserve inter-channel phase relations, which is key to accurate sound localization. Systematic evaluations using simulated and recorded room impulse responses demonstrate that the proposed model yields excellent frame-level speaker localization results in reverberant and noisy environments and outperforms related methods by a large margin, even surpassing their utterance-level results.

Deep learning-based speech enhancement (SE) methods often face significant computational challenges when needing to meet low-latency requirements because of the increased number of frames to be processed. This paper introduces the SlowFast framework which aims to reduce computation costs specifically when low-latency enhancement is needed. The framework consists of a slow branch that analyzes the acoustic environment at a low frame rate, and a fast branch that performs SE in the time domain at the needed higher frame rate to match the required latency. Specifically, the fast branch employs a state space model where its state transition process is dynamically modulated by the slow branch. Experiments on a SE task with a 2 ms algorithmic latency requirement using the Voice Bank + Demand dataset show that our approach reduces computation cost by 70% compared to a baseline single-branch network with equivalent parameters, without compromising enhancement performance. Furthermore, by leveraging the SlowFast framework, we implemented a network that achieves an algorithmic latency of just 62.5 μs (one sample point at 16 kHz sample rate) with a computation cost of 100 M MACs/s, while scoring a PESQ-NB of 3.12 and SISNR of 16.62.

This paper presents an improved approach to multimodal active speaker detection in egocentric videos, specifically designed to be robust against the rapid movements and motion blur commonly found in such videos. We propose two key techniques to improve the model’s resilience: (i) spatially fixing the lip region in the visual input, and (ii) applying motion blur augmentation. These methods significantly enhance the model’s performance in handling the challenges typical of egocentric videos. We showcase the effectiveness of these techniques on a simple but efficient causal audio-visual model. The proposed model, named EgoASD, demonstrates state-of-the-art performance on the EasyCom dataset, beating the previous SOTA by 1.7% mean Average Precision (mAP) with a model 2.5 times smaller. Our ablations highlight the importance of visual input, motion blur augmentation, the pretraining method and the importance of temporal context. To demonstrate its applicability in the real world, we apply our model to audio-visual speaker diarization, outperforming other baselines on EasyCom.

We present a streamlined framework for complex spectral masking that processes multichannel speech with minimal computational demands, enhancing both spectral magnitude and phase by integrating low-compute models with the Multi-Channel Wiener Filter (MCWF). Our methodology employs a two-stage, end-to-end training approach where a deep neural network (DNN) first estimates MCWF weights, followed by another DNN that refines the MCWF output, enhancing spectral masking quality. This architecture not only outperforms the traditional oracle Minimum Variance Distortionless Response (MVDR) beamformer but also maintains high efficiency, requiring less than 50MMACs for processing one second of 8-channel audio. Empirical results demonstrate that our framework exceeds the performance of existing low-compute models, offering significant enhancements with minimal computational demands, making it ideal for deployment on edge devices with limited computational resources.

MetricGAN, a notable generative approach, provides an effective framework to train speech enhancement models to produce high metric scores. However, we identify two key limitations of current MetricGAN-family models, i.e. neglecting certain mainstream metrics during evaluation and conducting evaluation exclusively at high SNR. Firstly, we comprehensively assess MetricGAN models using mainstream metrics, surprisingly revealing MetricGAN models produce worse SISDR and STOI than unprocessed noisy speech. Secondly, we demonstrate that training MetricGAN models at low SNR often results in convergence to biased local minima, where PESQ scores are inflated while their SISDR and STOI values deteriorate significantly. In addition, we propose and validate two training tricks to address these issues: SISDR regularization and mixture-of-actor training. We find that these tricks effectively guide MetricGAN models to avoid local minima, thus improving speech quality.

We present a novel model designed for resource-efficient multichannel speech enhancement in the time domain, with a focus on low latency, lightweight, and low computational requirements. The proposed model incorporates explicit spatial and temporal processing within deep neural network (DNN) layers. Inspired by frequency-dependent multichannel filtering, our spatial filtering process applies multiple trainable filters to each hidden unit across the spatial dimension, resulting in a multichannel output. The temporal processing is applied over a single-channel output stream from the spatial processing using a Long Short-Term Memory (LSTM) network. The output from the temporal processing stage is then further integrated into the spatial dimension through elementwise multiplication. This explicit separation of spatial and temporal processing results in a resource-efficient network design. Empirical findings from our experiments show that our proposed model significantly outperforms robust baseline models while demanding far fewer parameters and computations, while achieving an ultra-low algorithmic latency of just 2 milliseconds.

We introduce a time-domain framework for efficient multichannel speech enhancement, emphasizing low latency and computational efficiency. This framework incorporates two compact deep neural networks (DNNs) surrounding a multichannel neural Wiener filter (NWF). The first DNN enhances the speech signal to estimate NWF coefficients, while the second DNN refines the output from the NWF. The NWF, while conceptually similar to the traditional frequency-domain Wiener filter, undergoes a training process optimized for low-latency speech enhancement, involving fine-tuning of both analysis and synthesis transforms. Our research results illustrate that the NWF output, having minimal nonlinear distortions, attains performance levels akin to those of the first DNN, deviating from conventional Wiener filter paradigms. Training all components jointly outperforms sequential training, despite its simplicity. Consequently, this framework achieves superior performance with fewer parameters and reduced computational demands, making it a compelling solution for resource-efficient multichannel speech enhancement.

Continuous speaker separation aims to separate overlapping speakers in real-world environments like meetings, but it often falls short in isolating speech segments of a single speaker. This leads to split signals that adversely affect downstream applications such as automatic speech recognition and speaker diarization. Existing solutions like speaker counting have limitations. This paper presents a novel multi-channel approach for continuous speaker separation based on multi-input multi-output (MIMO) complex spectral mapping. This MIMO approach enables robust speaker localization by preserving inter-channel phase relations. Speaker localization as a byproduct of the MIMO separation model is then used to identify single-talker frames and reduce speaker splitting. We demonstrate that this approach achieves superior frame-level sound localization. Systematic experiments on the LibriCSS dataset further show that the proposed approach outperforms other methods, advancing state-of-the-art speaker separation performance.

This paper introduces a new Dynamic Gated Recurrent Neural Network (DG-RNN) for compute-efficient speech enhancement models running on resource-constrained hardware platforms.It leverages the slow evolution characteristic of RNN hidden states over steps, and updates only a selected set of neurons at each step by adding a newly proposed select gate to the RNN model.This select gate allows the computation cost of the conventional RNN to be reduced during network inference.As a realization of the DG-RNN, we further propose the Dynamic Gated Recurrent Unit (D-GRU) which does not require additional parameters.Test results obtained from several state-ofthe-art compute-efficient RNN-based speech enhancement architectures using the DNS challenge dataset, show that the D-GRU based model variants maintain similar speech intelligibility and quality metrics comparable to the baseline GRU based models even with an average 50% reduction in GRU computes.

It has been shown that the intelligibility of noisy speech can be improved by speech enhancement algorithms. However, speech enhancement has not been established as an effective frontend for robust automatic speech recognition (ASR) in noisy conditions compared to an ASR model trained on noisy speech directly. The divide between speech enhancement and ASR impedes the progress of robust ASR systems especially as speech enhancement has made big strides in recent years. In this work, we focus on eliminating this divide with an ARN (attentive recurrent network) based time-domain enhancement model. The proposed system fully decouples speech enhancement and an acoustic model trained only on clean speech. Results on the CHiME-2 corpus show that ARN enhanced speech translates to improved ASR results. The proposed system achieves 6.28% average word error rate, outperforming the previous best by 19.3% relatively.

Dealing with speech interference in a speech enhancement system requires either speaker separation or target speaker extraction. Speaker separation has multiple output streams with arbitrary assignments while target speaker extraction requires additional cueing for speaker selection. Both of these are not suitable for a standalone speech enhancement system with one output stream. In this study, we propose a novel training framework, calledAttentive Training, to extend speech enhancement to deal with speech interruptions. Attentive training is based on the observation that, in the real world, multiple talkers very unlikely start speaking at the same time, and therefore, a deep neural network can be trained to create a representation of the first speaker and utilize it to attend to or track that speaker in a multitalker noisy mixture. We present experimental results and comparisons to demonstrate the effectiveness of attentive training for speech enhancement.

Processing latency is a critical issue for active noise control (ANC) due to the causality constraint of ANC systems. This paper addresses low-latency ANC in the context of deep learning (i.e. deep ANC). A time-domain method using an attentive recurrent network (ARN) is employed to perform deep ANC with smaller frame sizes, thus reducing algorithmic latency of deep ANC. In addition, we introduce a delay-compensated training to perform ANC using predicted noise for several milliseconds. Moreover, a revised overlap-add method is utilized during signal resynthesis to avoid the latency introduced due to overlaps between neighboring time frames. Experimental results show the effectiveness of the proposed strategies for achieving low-latency deep ANC. Combining the proposed strategies is capable of yielding zero, even negative, algorithmic latency without affecting ANC performance much, thus alleviating the causality constraint in ANC design.

In this work, we propose a new model called triple-path attentive recurrent network (TPARN) for multichannel speech enhancement in the time domain. TPARN extends a single-channel dual-path network to a multichannel network by adding a third path along the spatial dimension. First, TPARN processes speech signals from all channels independently using a dual-path attentive recurrent network (ARN), which is a recurrent neural network (RNN) augmented with self-attention. Next, an ARN is introduced along the spatial dimension for spatial context aggregation. TPARN is designed as a multiple-input and multiple-output architecture to enhance all input channels simultaneously. Experimental results demonstrate the superiority of TPARN over existing state-of-the-art approaches.

Deep neural networks are often coupled with traditional spatial filters, such as MVDR beamformers for effectively exploiting spatial information. Even though single-stage end-to-end supervised models can obtain impressive enhancement, combining them with a traditional beamformer and a DNN-based post-filter in a multistage processing provides additional improvements. In this work, we propose a two-stage strategy for multi-channel speech enhancement that does not require a traditional beamformer for additional performance. First, we propose a novel attentive dense convolutional network (ADCN) for estimating real and imaginary parts of complex spectrogram. ADCN obtains state-of-the-art results among single-stage models. Next, we use ADCN with a recently proposed triple-path attentive recurrent network (TPARN) for estimating waveform samples. The proposed strategy uses two insights; first, using different approaches in two stages; and second, using a stronger model in the first stage. We illustrate the efficacy of our strategy by evaluating multiple models in a two-stage approach with and without a traditional beamformer.

Deep neural networks (DNNs) are very effective for multichannel speech enhancement with fixed array geometries.However, it is not trivial to use DNNs for ad-hoc arrays with unknown order and placement of microphones.We propose a novel triplepath network for ad-hoc array processing in the time domain.The key idea in the network design is to divide the overall processing into spatial processing and temporal processing and use self-attention for spatial processing.Using self-attention for spatial processing makes the network invariant to the order and the number of microphones.The temporal processing is done independently for all channels using a recently proposed dual-path attentive recurrent network.The proposed network is a multiple-input multiple-output architecture that can simultaneously enhance signals at all microphones.Experimental results demonstrate the excellent performance of the proposed approach.Further, we present analysis to demonstrate the effectiveness of the proposed network in utilizing multichannel information even from microphones at far locations.

Processing latency is a critical issue for active noise control (ANC) due to the causality constraint of ANC systems. This paper addresses low-latency ANC in the deep learning framework (i.e. deep ANC). A time-domain method using an attentive recurrent network is employed to perform deep ANC with smaller frame sizes, thus reducing algorithmic latency of deep ANC. In addition, a delay-compensated training strategy is introduced to perform ANC using predicted noise for several milliseconds. Moreover, we utilize a revised overlap-add method during signal resynthesis to avoid the latency introduced due to overlaps between neighboring time frames. Experimental results show that the proposed strategies are effective for achieving low-latency deep ANC. Combining the proposed strategies is capable of yielding zero, even negative, algorithmic latency without significantly affecting ANC performance.

Deep neural networks (DNNs) represent the mainstream methodology for supervised speech enhancement, primarily due to their capability to model complex functions using hierarchical representations. However, a recent study revealed that DNNs trained on a single corpus fail to generalize to untrained corpora, especially in low signal-to-noise ratio (SNR) conditions. Developing a noise, speaker, and corpus independent speech enhancement algorithm is essential for real-world applications. In this study, we propose a self-attending recurrent neural network (SARNN) for time-domain speech enhancement to improve cross-corpus generalization. SARNN comprises of recurrent neural networks (RNNs) augmented with self-attention blocks and feedforward blocks. We evaluate SARNN on different corpora with nonstationary noises in low SNR conditions. Experimental results demonstrate that SARNN substantially outperforms competitive approaches to time-domain speech enhancement, such as RNNs and dual-path SARNNs. Additionally, we report an important finding that the two popular approaches to speech enhancement: complex spectral mapping and time-domain enhancement, obtain similar results for RNN and SARNN with large-scale training. We also provide a challenging subset of the test set used in this study for evaluating future algorithms and facilitating direct comparisons.

In this work, we exploit speech enhancement for improving a re-current neural network transducer (RNN-T) based ASR system. We employ a dense convolutional recurrent network (DCRN) for complex spectral mapping based speech enhancement, and find it helpful for ASR in two ways: a data augmentation technique, and a preprocessing frontend. In using it for ASR data augmentation, we exploit a KL divergence based consistency loss that is computed between the ASR outputs of original and enhanced utterances. In using speech enhancement as an effective ASR frontend, we propose a three-step training scheme based on model pretraining and feature selection. We evaluate our proposed techniques on a challenging social media English video dataset, and achieve an average relative improvement of 11.2% with speech enhancement based data augmentation, 8.3% with enhancement based preprocessing, and 13.4% when combining both.

Speech enhancement in the time domain is becoming increasingly popular in recent years, due to its capability to jointly enhance both the magnitude and the phase of speech. In this work, we propose a dense convolutional network (DCN) with self-attention for speech enhancement in the time domain. DCN is an encoder and decoder based architecture with skip connections. Each layer in the encoder and the decoder comprises a dense block and an attention module. Dense blocks and attention modules help in feature extraction using a combination of feature reuse, increased network depth, and maximum context aggregation. Furthermore, we reveal previously unknown problems with a loss based on the spectral magnitude of enhanced speech. To alleviate these problems, we propose a novel loss based on magnitudes of enhanced speech and a predicted noise. Even though the proposed loss is based on magnitudes only, a constraint imposed by noise prediction ensures that the loss enhances both magnitude and phase. Experimental results demonstrate that DCN trained with the proposed loss substantially outperforms other state-of-the-art approaches to causal and non-causal speech enhancement.

In this work, we propose a fully convolutional neural network for real-time speech enhancement in the time domain. The proposed network is an encoder-decoder based architecture with skip connections. The layers in the encoder and the decoder are followed by densely connected blocks comprising of dilated and causal convolutions. The dilated convolutions help in context aggregation at different resolutions. The causal convolutions are used to avoid information flow from future frames, hence making the network suitable for real-time applications. We also propose to use sub-pixel convolutional layers in the decoder for upsampling. Further, the model is trained using a loss function with two components; a time-domain loss and a frequency-domain loss. The proposed loss function outperforms the time-domain loss. Experimental results show that the proposed model significantly outperforms other real-time state-of-the-art models in terms of objective intelligibility and quality scores.

In recent years, supervised approaches using deep neural networks (DNNs) have become the mainstream for speech enhancement. It has been established that DNNs generalize well to untrained noises and speakers if trained using a large number of noises and speakers. However, we find that DNNs fail to generalize to new speech corpora in low signal-to-noise ratio (SNR) conditions. In this work, we establish that the lack of generalization is mainly due to the channel mismatch, i.e. different recording conditions between the trained and untrained corpus. Additionally, we observe that traditional channel normalization techniques are not effective in improving cross-corpus generalization. Further, we evaluate publicly available datasets that are promising for generalization. We find one particular corpus to be significantly better than others. Finally, we find that using a smaller frame shift in short-time processing of speech can significantly improve cross-corpus generalization. The proposed techniques to address cross-corpus generalization include channel normalization, better training corpus, and smaller frame shift in short-time Fourier transform (STFT). These techniques together improve the objective intelligibility and quality scores on untrained corpora significantly.

This work proposes a fully convolutional neural network (CNN) for real-time speech enhancement in the time domain. The proposed CNN is an encoder-decoder based architecture with an additional temporal convolutional module (TCM) inserted between the encoder and the decoder. We call this architecture a Temporal Convolutional Neural Network (TCNN). The encoder in the TCNN creates a low dimensional representation of a noisy input frame. The TCM uses causal and dilated convolutional layers to utilize the encoder output of the current and previous frames. The decoder uses the TCM output to reconstruct the enhanced frame. The proposed model is trained in a speaker- and noise-independent way. Experimental results demonstrate that the proposed model gives consistently better enhancement results than a state-of-the-art real-time convolutional recurrent model. Moreover, since the model is fully convolutional, it has much fewer trainable parameters than earlier models.

A recent study has demonstrated the effectiveness of complex-valued deep neural networks (CDNNs) using newly developed tools such as complex batch normalization and complex residual blocks. Motivated by the fact that CDNNs are well suited for the processing of complex-domain representations, we explore CDNNs for speech enhancement. In particular, we train a CDNN that learns to map the complex-valued noisy short-time Fourier transform (STFT) to the clean STFT. Additionally, we propose the complex-valued extensions of the parametric rectified linear unit (PReLU) nonlinearity that helps to improve the performance of CDNN. Experimental results demonstrate that a CDNN using the proposed nonlinearity can give similar or better enhancement results compared to real-valued deep neural networks (DNNs).

This paper proposes a new learning mechanism for a fully convolutional neural network (CNN) to address speech enhancement in the time domain. The CNN takes as input the time frames of noisy utterance and outputs the time frames of the enhanced utterance. At the training time, we add an extra operation that converts the time domain to the frequency domain. This conversion corresponds to simple matrix multiplication, and is hence differentiable implying that a frequency domain loss can be used for training in the time domain. We use mean absolute error loss between the enhanced short-time Fourier transform (STFT) magnitude and the clean STFT magnitude to train the CNN. This way, the model can exploit the domain knowledge of converting a signal to the frequency domain for analysis. Moreover, this approach avoids the well-known invalid STFT problem since the proposed CNN operates in the time domain. Experimental results demonstrate that the proposed method substantially outperforms the other methods of speech enhancement. The proposed method is easy to implement and applicable to related speech processing tasks that require time-frequency masking or spectral mapping.

Generative adversarial networks (GANs) are becoming increasingly popular for image processing tasks. Researchers have started using GAN s for speech enhancement, but the advantage of using the GAN framework has not been established for speech enhancement. For example, a recent study reports encouraging enhancement results, but we find that the architecture of the generator used in the GAN gives better performance when it is trained alone using the L1loss. This work presents a new GAN for speech enhancement, and obtains performance improvement with the help of adversarial training. A deep neural network (DNN) is used for time-frequency mask estimation, and it is trained in two ways: regular training with the L1loss and training using the GAN framework with the help of an adversary discriminator. Experimental results suggest that the GAN framework improves speech enhancement performance. Further exploration of loss functions, for speech enhancement, suggests that the L1loss is consistently better than the L2loss for improving the perceptual quality of noisy speech.