Emiru Tsunoo

Siddhant Arora, Hayato Futami, Jee-weon Jung et al. · nvidia

Recent studies leverage large language models with multi-tasking capabilities, using natural language prompts to guide the model's behavior and surpassing performance of task-specific models. Motivated by this, we ask: can we build a single model that jointly performs various spoken language understanding (SLU) tasks? We start by adapting a pre-trained automatic speech recognition model to additional tasks using single-token task specifiers. We enhance this approach through instruction tuning, i.e., finetuning by describing the task using natural language instructions followed by the list of label options. Our approach can generalize to new task descriptions for the seen tasks during inference, thereby enhancing its user-friendliness. We demonstrate the efficacy of our single multi-task learning model "UniverSLU" for 12 speech classification and sequence generation task types spanning 17 datasets and 9 languages. On most tasks, UniverSLU achieves competitive performance and often even surpasses task-specific models. Additionally, we assess the zero-shot capabilities, finding that the model generalizes to new datasets and languages for seen task types.

Alkis Koudounas, Hayato Futami, Quentin Jodelet et al.

Speech-to-speech translation (S2ST) has advanced rapidly, but offline evaluation lacks a unified protocol: studies report non-overlapping metric subsets, preventing direct comparisons. We introduce COMPASS, a unified and reproducible benchmarking framework integrating 46 metrics across eight dimensions, and deploy it on 1,248 model-language configurations from FLEURS and CVSS, spanning cascaded and end-to-end architectures over ten language pairs. Architectures exhibit complementary strengths: best-vs-worst gaps exceed 30\% on naturalness and speaker preservation but remain within a few points on translation quality, so single-metric rankings systematically misrepresent system quality. Correlation filtering reduces 46 metrics to 10 per direction, with three axes requiring different metrics across X$\to$EN and EN$\to$X (e.g., TER/UTMOS vs. ChrF++/NISQA-MOS); these subsets preserve rankings (Spearman's $ρ>0.80$) while cutting evaluation time by $\approx 2.5\times$. Human validation across dubbing, podcasts, and medical domains shows standalone MOS predictors fail to predict listener preference, while top domain-specific metrics correlate with human judgment ($ρ\geq 0.90$). We release COMPASS as a foundation for domain-aware S2ST evaluation.

Hayato Futami, Emiru Tsunoo, Kentaro Shibata et al.

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 Switchboard and the corpus of spontaneous Japanese.

Siddhant Arora, Jinchuan Tian, Jiatong Shi et al.

Reinforcement learning from human or AI feedback (RLHF/RLAIF) for speech-in/speech-out dialogue systems (SDS) remains underexplored, with prior work largely limited to single semantic rewards applied at the utterance level. Such setups overlook the multi-dimensional and multi-modal nature of conversational quality, which encompasses semantic coherence, audio naturalness, speaker consistency, emotion alignment, and turn-taking behavior. Moreover, they are fundamentally mismatched with duplex spoken dialogue systems that generate responses incrementally, where agents must make decisions based on partial utterances. We address these limitations with the first multi-reward RLAIF framework for SDS, combining semantic, audio-quality, and emotion-consistency rewards. To align utterance-level preferences with incremental, blockwise decoding in duplex models, we apply turn-level preference sampling and aggregate per-block log-probabilities within a single DPO objective. We present the first systematic study of preference learning for improving SDS quality in both multi-turn Chain-of-Thought and blockwise duplex models, and release a multi-reward DPO dataset to support reproducible research. Experiments show that single-reward RLAIF selectively improves its targeted metric, while joint multi-reward training yields consistent gains across semantic quality and audio naturalness. These results highlight the importance of holistic, multi-reward alignment for practical conversational SDS.

Yao-Fei Cheng, Hayato Futami, Yosuke Kashiwagi et al.

Recent progress in large language models (LLMs) has gained interest in speech-text multimodal foundation models, achieving strong performance on instruction-tuned speech translation (ST). However, expanding language pairs is costly due to re-training on combined new and previous datasets. To address this, we aim to build a one-to-many ST system from existing one-to-one ST systems using task arithmetic without re-training. Direct application of task arithmetic in ST leads to language confusion; therefore, we introduce an augmented task arithmetic method incorporating a language control model to ensure correct target language generation. Our experiments on MuST-C and CoVoST-2 show BLEU score improvements of up to 4.66 and 4.92, with COMET gains of 8.87 and 11.83. In addition, we demonstrate our framework can extend to language pairs lacking paired ST training data or pre-trained ST models by synthesizing ST models based on existing machine translation (MT) and ST models via task analogies.

Siddhant Arora, Hayato Futami, Yosuke Kashiwagi et al.

There has been an increased interest in the integration of pretrained speech recognition (ASR) and language models (LM) into the SLU framework. However, prior methods often struggle with a vocabulary mismatch between pretrained models, and LM cannot be directly utilized as they diverge from its NLU formulation. In this study, we propose a three-pass end-to-end (E2E) SLU system that effectively integrates ASR and LM subnetworks into the SLU formulation for sequence generation tasks. In the first pass, our architecture predicts ASR transcripts using the ASR subnetwork. This is followed by the LM subnetwork, which makes an initial SLU prediction. Finally, in the third pass, the deliberation subnetwork conditions on representations from the ASR and LM subnetworks to make the final prediction. Our proposed three-pass SLU system shows improved performance over cascaded and E2E SLU models on two benchmark SLU datasets, SLURP and SLUE, especially on acoustically challenging utterances.

Yosuke Kashiwagi, Hayato Futami, Emiru Tsunoo et al.

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.

Siddhant Arora, Yifan Peng, Jiatong Shi et al. · nvidia

Advancements in audio foundation models (FMs) have fueled interest in end-to-end (E2E) spoken dialogue systems, but different web interfaces for each system makes it challenging to compare and contrast them effectively. Motivated by this, we introduce an open-source, user-friendly toolkit designed to build unified web interfaces for various cascaded and E2E spoken dialogue systems. Our demo further provides users with the option to get on-the-fly automated evaluation metrics such as (1) latency, (2) ability to understand user input, (3) coherence, diversity, and relevance of system response, and (4) intelligibility and audio quality of system output. Using the evaluation metrics, we compare various cascaded and E2E spoken dialogue systems with a human-human conversation dataset as a proxy. Our analysis demonstrates that the toolkit allows researchers to effortlessly compare and contrast different technologies, providing valuable insights such as current E2E systems having poorer audio quality and less diverse responses. An example demo produced using our toolkit is publicly available here: https://huggingface.co/spaces/Siddhant/Voice_Assistant_Demo.

Hayato Futami, Emiru Tsunoo, Yosuke Kashiwagi et al.

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, Jinchuan Tian, Hayato Futami et al.

Unlike traditional cascaded pipelines, end-to-end (E2E) spoken dialogue systems preserve full differentiability and capture non-phonemic information, making them well-suited for modeling spoken interactions. However, existing E2E approaches often require large-scale training data and generates responses lacking semantic coherence. We propose a simple yet effective strategy leveraging a chain-of-thought (CoT) formulation, ensuring that training on conversational data remains closely aligned with the multimodal language model (LM)'s pre-training on speech recognition~(ASR), text-to-speech synthesis (TTS), and text LM tasks. Our method achieves over 1.5 ROUGE-1 improvement over the baseline, successfully training spoken dialogue systems on publicly available human-human conversation datasets, while being compute-efficient enough to train on just 300 hours of public human-human conversation data, such as the Switchboard. We will publicly release our models and training code.

Yosuke Kashiwagi, Hayato Futami, Emiru Tsunoo et al.

This paper reports on the development of a large-scale speech recognition model, Whale. Similar to models such as Whisper and OWSM, Whale leverages both a large model size and a diverse, extensive dataset. Whale's architecture integrates w2v-BERT self-supervised model, an encoder-decoder backbone built on E-Branchformer, and a joint CTC-attention decoding strategy. The training corpus comprises varied speech data, of not only public corpora but also in-house data, thereby enhancing the model's robustness to different speaking styles and acoustic conditions. Through evaluations on multiple benchmarks, Whale achieved comparable performance to existing models. In particular, it achieves a word error rate of 2.4% on the Librispeech test-clean set and a character error rate of 3.4% on the CSJ eval3 set, outperforming Whisper large-v3 and OWSM v3.1.

Wataru Nakata, Yuki Saito, Kazuki Yamauchi et al.

Full-duplex dialogue audio, in which each speaker is recorded on a separate track, is an important resource for spoken dialogue research, but is difficult to collect at scale. Most in-the-wild two-speaker dialogue is available only as degraded monaural mixtures, making it unsuitable for systems requiring clean speaker-wise signals. We propose DialogueSidon, a model for joint restoration and separation of degraded monaural two-speaker dialogue audio. DialogueSidon combines a variational autoencoder (VAE) operates on the speech self-supervised learning (SSL) model feature, which compresses SSL model features into a compact latent space, with a diffusion-based latent predictor that recovers speaker-wise latent representations from the degraded mixture. Experiments on English, multilingual, and in-the-wild dialogue datasets show that DialogueSidon substantially improves intelligibility and separation quality over a baseline, while also achieving much faster inference.

Siddhant Arora, Jinchuan Tian, Hayato Futami et al.

Most end-to-end (E2E) spoken dialogue systems (SDS) rely on voice activity detection (VAD) for turn-taking, but VAD fails to distinguish between pauses and turn completions. Duplex SDS models address this by predicting output continuously, including silence tokens, thus removing the need for explicit VAD. However, they often have complex dual-channel architecture and lag behind cascaded models in semantic reasoning. To overcome these challenges, we propose SCoT: a Streaming Chain-of-Thought (CoT) framework for Duplex SDS, alternating between processing fixed-duration user input and generating responses in a blockwise manner. Using frame-level alignments, we create intermediate targets-aligned user transcripts and system responses for each block. Experiments show that our approach produces more coherent and interpretable responses than existing duplex methods while supporting lower-latency and overlapping interactions compared to turn-by-turn systems.

Emiru Tsunoo, Hayato Futami, Yosuke Kashiwagi et al.

For streaming speech recognition, a Transformer-based 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.

Hayato Futami, Emiru Tsunoo, Yosuke Kashiwagi et al.

Speech-to-speech translation (S2ST) has been advanced with large language models (LLMs), which are fine-tuned on discrete speech units. In such approaches, modality adaptation from text to speech has been an issue. LLMs are trained on text-only data, which presents challenges to adapt them to speech modality with limited speech-to-speech data. To address the training difficulty, we propose scheduled interleaved speech--text training in this study. We use interleaved speech--text units instead of speech units during training, where aligned text tokens are interleaved at the word level. We gradually decrease the ratio of text as training progresses, to facilitate progressive modality adaptation from text to speech. We conduct experimental evaluations by fine-tuning LLaMA3.2-1B for S2ST on the CVSS dataset. We show that the proposed method consistently improves the translation performances, especially for languages with limited training data.

Emiru Tsunoo, Hayato Futami, Yosuke Kashiwagi et al.

Decoder-only language models (LMs) have been successfully adopted for speech-processing tasks including automatic speech recognition (ASR). The LMs have ample expressiveness and perform efficiently. This efficiency is a suitable characteristic for streaming applications of ASR. In this work, we propose to use a decoder-only architecture for blockwise streaming ASR. In our approach, speech features are compressed using CTC output and context embedding using blockwise speech subnetwork, and are sequentially provided as prompts to the decoder. The decoder estimates the output tokens promptly at each block. To this end, we also propose a novel training scheme using random-length prefix prompts to make the model robust to the truncated prompts caused by blockwise processing. An experimental comparison shows that our proposed decoder-only streaming ASR achieves 8% relative word error rate reduction in the LibriSpeech test-other set while being twice as fast as the baseline model.

Yosuke Kashiwagi, Hayato Futami, Emiru Tsunoo et al.

End-to-end multilingual speech recognition models handle multiple languages through a single model, often incorporating language identification to automatically detect the language of incoming speech. Since the common scenario is where the language is already known, these models can perform as language-specific by using language information as prompts, which is particularly beneficial for attention-based encoder-decoder architectures. However, the Connectionist Temporal Classification (CTC) approach, which enhances recognition via joint decoding and multi-task training, does not normally incorporate language prompts due to its conditionally independent output tokens. To overcome this, we introduce an encoder prompting technique within the self-conditioned CTC framework, enabling language-specific adaptation of the CTC model in a zero-shot manner. Our method has shown to significantly reduce errors by 28% on average and by 41% on low-resource languages.

Hayato Futami, Siddhant Arora, Yosuke Kashiwagi et al.

Recently, multi-task spoken language understanding (SLU) models have emerged, designed to address various speech processing tasks. However, these models often rely on a large number of parameters. Also, they often encounter difficulties in adapting to new data for a specific task without experiencing catastrophic forgetting of previously trained tasks. In this study, we propose finding task-specific subnetworks within a multi-task SLU model via neural network pruning. In addition to model compression, we expect that the forgetting of previously trained tasks can be mitigated by updating only a task-specific subnetwork. We conduct experiments on top of the state-of-the-art multi-task SLU model ``UniverSLU'', trained for several tasks such as emotion recognition (ER), intent classification (IC), and automatic speech recognition (ASR). We show that pruned models were successful in adapting to additional ASR or IC data with minimal performance degradation on previously trained tasks.

Siddhant Arora, Hayato Futami, Shih-Lun Wu et al.

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.

Hayato Futami, Jessica Huynh, Siddhant Arora et al.

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.

Siddhant Arora, Hayato Futami, Emiru Tsunoo et al.

Most human interactions occur in the form of spoken conversations where the semantic meaning of a given utterance depends on the context. Each utterance in spoken conversation can be represented by many semantic and speaker attributes, and there has been an interest in building Spoken Language Understanding (SLU) systems for automatically predicting these attributes. Recent work has shown that incorporating dialogue history can help advance SLU performance. However, separate models are used for each SLU task, leading to an increase in inference time and computation cost. Motivated by this, we aim to ask: can we jointly model all the SLU tasks while incorporating context to facilitate low-latency and lightweight inference? To answer this, we propose a novel model architecture that learns dialog context to jointly predict the intent, dialog act, speaker role, and emotion for the spoken utterance. Note that our joint prediction is based on an autoregressive model and we need to decide the prediction order of dialog attributes, which is not trivial. To mitigate the issue, we also propose an order agnostic training method. Our experiments show that our joint model achieves similar results to task-specific classifiers and can effectively integrate dialog context to further improve the SLU performance.

Chaitanya Narisetty, Emiru Tsunoo, Xuankai Chang et al.

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.

Emiru Tsunoo, Chaitanya Narisetty, Michael Hentschel et al.

A streaming style inference of encoder-decoder automatic speech recognition (ASR) system 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, endpoint 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.

Rem Hida, Masaki Hamada, Chie Kamada et al.

Although end-to-end text-to-speech (TTS) models can generate natural speech, challenges still remain when it comes to estimating sentence-level phonetic and prosodic information from raw text in Japanese TTS systems. In this paper, we propose a method for polyphone disambiguation (PD) and accent prediction (AP). The proposed method incorporates explicit features extracted from morphological analysis and implicit features extracted from pre-trained language models (PLMs). We use BERT and Flair embeddings as implicit features and examine how to combine them with explicit features. Our objective evaluation results showed that the proposed method improved the accuracy by 5.7 points in PD and 6.0 points in AP. Moreover, the perceptual listening test results confirmed that a TTS system employing our proposed model as a front-end achieved a mean opinion score close to that of synthesized speech with ground-truth pronunciation and accent in terms of naturalness.

Kazuki Shimada, Yuichiro Koyama, Shusuke Takahashi et al.

Sound event localization and detection (SELD) involves identifying the direction-of-arrival (DOA) and the event class. The SELD methods with a class-wise output format make the model predict activities of all sound event classes and corresponding locations. The class-wise methods can output activity-coupled Cartesian DOA (ACCDOA) vectors, which enable us to solve a SELD task with a single target using a single network. However, there is still a challenge in detecting the same event class from multiple locations. To overcome this problem while maintaining the advantages of the class-wise format, we extended ACCDOA to a multi one and proposed auxiliary duplicating permutation invariant training (ADPIT). The multi- ACCDOA format (a class- and track-wise output format) enables the model to solve the cases with overlaps from the same class. The class-wise ADPIT scheme enables each track of the multi-ACCDOA format to learn with the same target as the single-ACCDOA format. In evaluations with the DCASE 2021 Task 3 dataset, the model trained with the multi-ACCDOA format and with the class-wise ADPIT detects overlapping events from the same class while maintaining its performance in the other cases. Also, the proposed method performed comparably to state-of-the-art SELD methods with fewer parameters.

Yuichiro Koyama, Kazuhide Shigemi, Masafumi Takahashi et al.

Recording and annotating real sound events for a sound event localization and detection (SELD) task is time consuming, and data augmentation techniques are often favored when the amount of data is limited. However, how to augment the spatial information in a dataset, including unlabeled directional interference events, remains an open research question. Furthermore, directional interference events make it difficult to accurately extract spatial characteristics from target sound events. To address this problem, we propose an impulse response simulation framework (IRS) that augments spatial characteristics using simulated room impulse responses (RIR). RIRs corresponding to a microphone array assumed to be placed in various rooms are accurately simulated, and the source signals of the target sound events are extracted from a mixture. The simulated RIRs are then convolved with the extracted source signals to obtain an augmented multi-channel training dataset. Evaluation results obtained using the TAU-NIGENS Spatial Sound Events 2021 dataset show that the IRS contributes to improving the overall SELD performance. Additionally, we conducted an ablation study to discuss the contribution and need for each component within the IRS.

Kazuki Shimada, Naoya Takahashi, Yuichiro Koyama et al.

This report describes our systems submitted to the DCASE2021 challenge task 3: sound event localization and detection (SELD) with directional interference. Our previous system based on activity-coupled Cartesian direction of arrival (ACCDOA) representation enables us to solve a SELD task with a single target. This ACCDOA-based system with efficient network architecture called RD3Net and data augmentation techniques outperformed state-of-the-art SELD systems in terms of localization and location-dependent detection. Using the ACCDOA-based system as a base, we perform model ensembles by averaging outputs of several systems trained with different conditions such as input features, training folds, and model architectures. We also use the event independent network v2 (EINV2)-based system to increase the diversity of the model ensembles. To generalize the models, we further propose impulse response simulation (IRS), which generates simulated multi-channel signals by convolving simulated room impulse responses (RIRs) with source signals extracted from the original dataset. Our systems significantly improved over the baseline system on the development dataset.

Emiru Tsunoo, Kentaro Shibata, Chaitanya Narisetty et al.

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 distant-talk 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) cycle-consistent 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.

Yosuke Kashiwagi, Emiru Tsunoo, Shinji Watanabe

Self-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.

Emiru Tsunoo, Yosuke Kashiwagi, Shinji Watanabe

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.

Emiru Tsunoo, Yosuke Kashiwagi, Toshiyuki Kumakura et al.

The Transformer self-attention network has recently 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 self-attention. We have proposed a 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. In this paper, we extend it towards an entire online E2E ASR system by introducing an online decoding process inspired by monotonic chunkwise attention (MoChA) into the Transformer decoder. Our novel MoChA training and inference algorithms exploit the unique properties of Transformer, whose attentions are not always monotonic or peaky, and have multiple heads and residual connections of the decoder layers. Evaluations of the Wall Street Journal (WSJ) and AISHELL-1 show that our proposed online Transformer decoder outperforms conventional chunkwise approaches.

Emiru Tsunoo, Yosuke Kashiwagi, Toshiyuki Kumakura et al.

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.

Emiru Tsunoo, Yosuke Kashiwagi, Satoshi Asakawa et al.

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.