For streaming speech recognition, a Transformerbased encoder has been widely used with block processing. Although many studies addressed improving emission latency of transducers, little work has been explored for improving encoding latency of the block processing. We seek to reduce latency by frequently emitting a chunk with a small shift rather than scarce large-chunk emissions, resulting in higher computational costs. To efficiently compute with the small chunk shift, we propose a new encoder, Spiralformer, tailored for block processing by combining layer dropping and early exiting. We skip layer computation in a cyclic manner and shift the computed layer in each block spirally, which completes computation for all the layers over the block processing. Experimentally, we observed that our method achieved 21.6% reduction in the averaged token emission delay in Librispeech, and 7.0% in CSJ, compared with the baseline with similar computational cost and word error rates.

In many real-world scenarios, such as meetings, multiple speakers are present with an unknown number of participants, and their utterances often overlap. We address these multi-speaker challenges by a novel attention-based encoder-decoder method augmented with special speaker class tokens obtained by speaker clustering. During inference, we select multiple recognition hypotheses conditioned on predicted speaker cluster tokens, and these hypotheses are merged by agglomerative hierarchical clustering (AHC) based on the normalized edit distance. The clustered hypotheses result in the multi-speaker transcriptions with the appropriate number of speakers determined by AHC. Our experiments on the LibriMix dataset demonstrate that our proposed method was particularly effective in complex 3-mix environments, achieving a 55% relative error reduction on clean data and a 36% relative error reduction on noisy data compared with conventional serialized output training.

In speech recognition applications, it is important to recognize context-specific rare words, such as proper nouns. Tree-constrained Pointer Generator (TCPGen) has shown promise for this purpose, which efficiently biases such words with a prefix tree. While the original TCPGen relies on grapheme-based encoding, we propose extending it with phoneme-aware encoding to better recognize words of unusual pronunciations. As TCPGen handles biasing words as subword units, we propose obtaining subword-level phoneme-aware encoding by using alignment between phonemes and subwords. Furthermore, we propose injecting phoneme-level predictions from CTC into queries of TCPGen so that the model better interprets the phoneme-aware encodings. We conducted ASR experiments with TCPGen for RNN transducer. We observed that proposed phoneme-aware encoding outperformed ordinary grapheme-based encoding on both the English LibriSpeech and Japanese CSJ datasets, demonstrating the robustness of our approach across linguistically diverse languages.

Siddhant Arora, Hayato Futami, Jee-weon Jung, Yifan Peng, Roshan Sharma, Yosuke Kashiwagi, Emiru Tsunoo, Karen Livescu, Shinji Watanabe. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Recently there have been efforts to introduce new benchmark tasks for spoken language understanding (SLU), like semantic parsing. In this paper, we describe our proposed spoken semantic parsing system for the quality track (Track 1) in Spoken Language Understanding Grand Challenge which is part of ICASSP Signal Processing Grand Challenge 2023. We experiment with both end-to-end and pipeline systems for this task. Strong automatic speech recognition (ASR) models like Whisper and pretrained Language models (LM) like BART are utilized inside our SLU framework to boost performance. We also investigate the output level combination of various models to get an exact match accuracy of 80.8, which won the 1st place at the challenge.

This paper describes our system for the low-resource domain adaptation track (Track 3) in Spoken Language Understanding Grand Challenge, which is a part of ICASSP Signal Processing Grand Challenge 2023. In the track, we adopt a pipeline approach of ASR and NLU. For ASR, we fine-tune Whisper for each domain with upsampling. For NLU, we fine-tune BART on all the Track3 data and then on low-resource domain data. We apply masked LM (MLM) -based data augmentation, where some of input tokens and corresponding target labels are replaced using MLM. We also apply a retrieval-based approach, where model input is augmented with similar training samples. As a result, we achieved exact match (EM) accuracy 63.3/75.0 (average: 69.15) for reminder/weather domain, and won the 1st place at the challenge.

Disfluency detection has mainly been solved in a pipeline approach, as post-processing of speech recognition. In this study, we propose Transformer-based encoder-decoder models that jointly solve speech recognition and disfluency detection, which work in a streaming manner. Compared to pipeline approaches, the joint models can leverage acoustic information that makes disfluency detection robust to recognition errors and provide non-verbal clues. Moreover, joint modeling results in low-latency and lightweight inference. We investigate two joint model variants for streaming disfluency detection: a transcript-enriched model and a multi-task model. The transcript- enriched model is trained on text with special tags indicating the starting and ending points of the disfluent part. However, it has problems with latency and standard language model adaptation, which arise from the additional disfluency tags. We propose a multi-task model to solve such problems, which has two output layers at the Transformer decoder; one for speech recognition and the other for disfluency detection. It is modeled to be conditioned on the currently recognized token with an additional token-dependency mechanism. We show that the proposed joint models outperformed a BERT-based pipeline approach in both accuracy and latency, on both the Switch- board and the corpus of spontaneous Japanese.

In this paper, we report our team’s study on track 2 of the Spoken Language Understanding Grand Challenge, which is a component of the ICASSP Signal Processing Grand Challenge 2023. The task is intended for on-device processing and involves estimating semantic parse labels from speech using a model with 15 million parameters. We use E2E E-Branchformer-based spoken language understanding model, which is more parameter controllable than cascade models, and reduced the parameter size through sequential distillation and tensor decomposition techniques. On the STOP dataset, we achieved an exact match accuracy of 70.9% under the tight constraint of 15 million parameters.

Speech samples recorded in both indoor and outdoor environments are often contaminated with secondary audio sources. Most end-to-end monaural speech recognition systems either remove these background sounds using speech enhancement or train noise-robust models. For better model interpretability and holistic understanding, we aim to bring together the growing field of automated audio captioning (AAC) and the thoroughly studied automatic speech recognition (ASR). The goal of AAC is to generate natural language descriptions of contents in audio samples. We propose several approaches for end-to-end joint modeling of ASR and AAC tasks and demonstrate their advantages over traditional approaches, which model these tasks independently. A major hurdle in evaluating our proposed approach is the lack of labeled audio datasets with both speech transcriptions and audio captions. Therefore we also create a multi-task dataset by mixing the clean speech Wall Street Journal corpus with multiple levels of background noises chosen from the AudioCaps dataset. We also perform extensive experimental evaluation and show improvements of our proposed methods as compared to existing state-of-the-art ASR and AAC methods.

A deep neural network (DNN)-based speech enhancement (SE) aiming to maximize the performance of an automatic speech recognition (ASR) system is proposed in this paper. In order to optimize the DNN-based SE model in terms of the character error rate (CER), which is one of the metric to evaluate the ASR system and generally non-differentiable, our method uses two DNNs: one for speech processing and one for mimicking the output CERs derived through an acoustic model (AM). Then both of DNNs are alternately optimized in the training phase. Even if the AM is a black-box, e.g., like one provided by a third-party, the proposed method enables the DNN-based SE model to be optimized in terms of the CER since the DNN mimicking the AM is differentiable. Consequently, it becomes feasible to build CER-centric SE model that has no negative effect, e.g., additional calculation cost and changing network architecture, on the inference phase since our method is merely a training scheme for the existing DNN-based methods. Experimental results show that our method improved CER by 8.8% relative derived through a black-box AM although certain noise levels are kept.

A streaming style inference of encoder–decoder automatic speech recognition (ASR) systems is important for reducing latency, which is essential for interactive use cases. To this end, we propose a novel blockwise synchronous decoding algorithm with a hybrid approach that combines endpoint prediction and endpoint post-determination. In the endpoint prediction, we compute the expectation of the number of tokens that are yet to be emitted in the encoder features of the current blocks using the CTC posterior. Based on the expectation value, the decoder predicts the endpoint to realize continuous block synchronization, as a running stitch. Meanwhile, end-point post-determination probabilistically detects backward jump of the source–target attention, which is caused by the misprediction of endpoints. Then it resumes decoding by discarding those hypotheses, as back stitch. We combine these methods into a hybrid approach, namely run-and-back stitch search, which reduces the computational cost and latency. Evaluations of various ASR tasks show the efficiency of our proposed decoding algorithm, which achieves a latency reduction, for instance in the Librispeech test set from 1487 ms to 821 ms at the 90th percentile, while maintaining a high recognition accuracy.

End-to-end automatic speech recognition suffers from adaptation to unknown target domain speech despite being trained with a large amount of paired audio-text data.Recent studies estimate a linguistic bias of the model as the internal language model (LM).To effectively adapt to the target domain, the internal LM is subtracted from the posterior during inference and fused with an external target-domain LM.However, this fusion complicates the inference and the estimation of the internal LM may not always be accurate.In this paper, we propose a simple external LM fusion method for domain adaptation, which considers the internal LM estimation in its training.We directly model the residual factor of the external and internal LMs, namely the residual LM.To stably train the residual LM, we propose smoothing the estimated internal LM and optimizing it with a combination of cross-entropy and mean-squared-error losses, which consider the statistical behaviors of the internal LM in the target domain data.We experimentally confirmed that the proposed residual LM performs better than the internal LM estimation in most of the cross-domain and intra-domain scenarios.

ISelf-attention (SA) based models have recently achieved significant performance improvements in hybrid and end-to-end automatic speech recognition (ASR) systems owing to their flexible context modeling capability. However, it is also known that the accuracy degrades when applying SA to long sequence data. This is mainly due to the length mismatch between the inference and training data because the training data are usually divided into short segments for efficient training. To mitigate this mismatch, we propose a new architecture, which is a variant of the Gaussian kernel, which itself is a shift-invariant kernel. First, we mathematically demonstrate that self-attention with shared weight parameters for queries and keys is equivalent to a normalized kernel function. By replacing this kernel function with the proposed Gaussian kernel, the architecture becomes completely shift-invariant with the relative position information embedded using a frame indexing technique. The proposed Gaussian kernelized SA was applied to connectionist temporal classification (CTC) based ASR. An experimental evaluation with the Corpus of Spontaneous Japanese (CSJ) and TEDLIUM 3 benchmarks shows that the proposed SA achieves a significant improvement in accuracy (e.g., from 24.0% WER to 6.0% in CSJ) in long sequence data without any windowing techniques.

Although end-to-end automatic speech recognition (E2E ASR) has achieved great performance in tasks that have numerous paired data, it is still challenging to make E2E ASR robust against noisy and low-resource conditions.In this study, we investigated data augmentation methods for E2E ASR in distanttalk scenarios.E2E ASR models are trained on the series of CHiME challenge datasets, which are suitable tasks for studying robustness against noisy and spontaneous speech.We propose to use three augmentation methods and thier combinations: 1) data augmentation using text-to-speech (TTS) data, 2) cycleconsistent generative adversarial network (Cycle-GAN) augmentation trained to map two different audio characteristics, the one of clean speech and of noisy recordings, to match the testing condition, and 3) pseudo-label augmentation provided by the pretrained ASR module for smoothing label distributions.Experimental results using the CHiME-6/CHiME-4 datasets show that each augmentation method individually improves the accuracy on top of the conventional SpecAugment; further improvements are obtained by combining these approaches.We achieved 4.3% word error rate (WER) reduction, which was more significant than that of the SpecAugment, when we combine all three augmentations for the CHiME-6 task.

The Transformer self-attention network has shown promising performance as an alternative to recurrent neural networks in end-to-end (E2E) automatic speech recognition (ASR) systems. However, Transformer has a drawback in that the entire input sequence is required to compute both self-attention and source-target attention. In this paper, we propose a novel blockwise synchronous beam search algorithm based on blockwise processing of encoder to perform streaming E2E Transformer ASR. In the beam search, encoded feature blocks are synchronously aligned using a block boundary detection technique, where a reliability score of each predicted hypothesis is evaluated based on the end-of-sequence and repeated tokens in the hypothesis. Evaluations of the HKUST and AISHELL-1 Mandarin, LibriSpeech English, and CSJ Japanese tasks show that the proposed streaming Transformer algorithm outperforms conventional online approaches, including monotonic chunkwise attention (MoChA), especially when using the knowledge distillation technique. An ablation study indicates that our streaming approach contributes to reducing the response time, and the repetition criterion contributes significantly in certain tasks. Our streaming ASR models achieve comparable or superior performance to batch models and other streaming-based Transformer methods in all tasks considered.

The Transformer self-attention network has recently shown promising performance as an alternative to recurrent neural networks (RNNs) in end-to-end (E2E) automatic speech recognition (ASR) systems. However, the Transformer has a drawback in that the entire input sequence is required to compute self-attention. In this paper, we propose a new block processing method for the Transformer encoder by introducing a context-aware inheritance mechanism. An additional context embedding vector handed over from the previously processed block helps to encode not only local acoustic information but also global linguistic, channel, and speaker attributes. We introduce a novel mask technique to implement the context inheritance to train the model efficiently. Evaluations of the Wall Street Journal (WSJ), Librispeech, VoxForge Italian, and AISHELL-1 Mandarin speech recognition datasets show that our proposed contextual block processing method outperforms naive block processing consistently. Furthermore, the attention weight tendency of each layer is analyzed to clarify how the added contextual inheritance mechanism models the global information.

An on-device DNN-HMM speech recognition system efficiently works with a limited vocabulary in the presence of a variety of predictable noise. In such a case, vocabulary and environment adaptation is highly effective. In this paper, we propose a novel method of end-to-end (E2E) adaptation, which adjusts not only an acoustic model (AM) but also a weighted finite-state transducer (WFST). We convert a pretrained WFST to a trainable neural network and adapt the system to target environments/vocabulary by E2E joint training with an AM. We replicate Viterbi decoding with forward--backward neural network computation, which is similar to recurrent neural networks (RNNs). By pooling output score sequences, a vocabulary posterior for each utterance is obtained and used for discriminative loss computation. Experiments using 2--10 hours of English/Japanese adaptation datasets indicate that the fine-tuning of only WFSTs and that of only AMs are both comparable to a state-of-the-art adaptation method, and E2E joint training of the two components achieves the best recognition performance. We also adapt each language system to the other language using the adaptation data, and the results show that the proposed method also works well for language adaptations.

In this paper, we propose a method to estimate statistical divergence between probability distributions by a DNN-based discriminative approach and its use for language identification tasks. Since statistical divergence is generally defined as a functional of two probability density functions, these density functions are usually represented in a parametric form. Then, if a mismatch exists between the assumed distribution and its true one, the obtained divergence becomes erroneous. In our proposed method, by using Bayes' theorem, the statistical divergence is estimated by using DNN as discriminative estimation model. In our method, the divergence between two distributions is able to be estimated without assuming a specific form for these distributions. When the amount of data available for estimation is small, however, it becomes intractable to calculate the integral of the divergence function over all the feature space and to train neural networks. To mitigate this problem, two solutions are introduced; a model adaptation method for DNN and a sampling approach for integration. We apply this approach to language identification tasks, where the obtained divergences are used to extract a speech structure. Experimental results show that our approach can improve the performance of language identification by 10.85% relative compared to the conventional approach based on i-vector.

The exemplar-based approaches, which model signals as a sparse linear combination of exemplars of signals, are proved to have state-of-the-art performance in noise robust ASR, especially on low SNRs. However, since both the speech exemplars and noise exemplars are built from training data and are fixed throughout the process of enhancing speech features, the conventional approach is especially weak for unknown types of noise. Therefore, in this paper, we propose a semi-supervised approach which automatically adapt noise exemplars to the target noise, while keeping the speech exemplars fixed. Continuous digits recognition experiments show that this approach is much more robust for unknown noise. The recognition errors are reduced by 36.2%.

In this paper, we propose the use of deep neural networks to expand conventional methods of statistical feature enhancement based on piecewise linear transformation. Stereo-based piecewise linear compensation for environments (SPLICE), which is a powerful statistical approach for feature enhancement, models the probabilistic distribution of input noisy features as a mixture of Gaussians. However, soft assignment of an input vector to divided regions is sometimes done inadequately and the vector comes to go through inadequate conversion. Especially when conversion has to be linear, the conversion performance will be easily degraded. Feature enhancement using neural networks is another powerful approach which can directly model a non-linear relationship between noisy and clean feature spaces. In this case, however, it tends to suffer from over-fitting problems. In this paper, we attempt to mitigate this problem by reducing the number of model parameters to estimate. Our neural network is trained whose output layer is associated with the states in the clean feature space, not in the noisy feature space. This strategy makes the size of the output layer independent of the kind of a given noisy environment. Firstly, we characterize the distribution of clean features as a Gaussian mixture model and then, by using deep neural networks, estimate discriminatively the state in the clean space that an input noisy feature corresponds to. Experimental evaluations using the Aurora 2 dataset demonstrate that our proposed method has the best performance compared to conventional methods.

Multimodal speech recognition is a promising approach to realize noise robust automatic speech recognition (ASR), and is currently gathering the attention of many researchers. Multimodal ASR utilizes not only audio features, which are sensitive to background noises, but also non-audio features such as lip shapes to achieve noise robustness. Although various methods have been proposed to integrate audio-visual features, there are still continuing discussions on how the vest integration of audio and visual features is realized. Weights of audio and visual features should be decided according to the noise features and levels: in general, larger weights to visual features when the noise level is low and vice versa, but how it can be controlled? In this paper, we propose a method based on piecewise linear transformation in feature integration. In contrast to other feature integration methods, our proposed method can appropriately change the weight depending on a state of an observed noisy feature, which has information both on uttered phonemes and environmental noise. Experiments on noisy speech recognition are conducted following to CENSREC-1-AV, and word error reduction rate around 24% is realized in average as compared to a decision fusion method.