Effectively distinguishing the pronunciation correlations between different written texts is a significant issue in linguistic acoustics. Traditionally, such pronunciation correlations are obtained through manually designed pronunciation lexicons. In this paper, we propose a data-driven method to automatically acquire these pronunciation correlations, called automatic text pronunciation correlation (ATPC). The supervision required for this method is consistent with the supervision needed for training end-to-end automatic speech recognition (E2E-ASR) systems, i.e., speech and corresponding text annotations. First, the iteratively-trained timestamp estimator (ITSE) algorithm is employed to align the speech with their corresponding annotated text symbols. Then, a speech encoder is used to convert the speech into speech embeddings. Finally, we compare the speech embeddings distances of different text symbols to obtain ATPC. Experimental results on Mandarin show that ATPC enhances E2E-ASR performance in contextual biasing and holds promise for dialects or languages lacking artificial pronunciation lexicons.

Recently, "textless" speech language models (SLMs) based on speech units have made huge progress in generating naturalistic speech, including non-verbal vocalizations. However, the generated speech samples often lack semantic coherence. In this paper, we propose SLM and LLM Integration for spontaneous spoken Dialogue gEneration (SLIDE). Specifically, we first utilize an LLM to generate the textual content of spoken dialogue. Next, we convert the textual dialogues into phoneme sequences and use a two-tower transformer-based duration predictor to predict the duration of each phoneme. Finally, an SLM conditioned on the spoken phoneme sequences is used to vocalize the textual dialogue. Experimental results on the Fisher dataset demonstrate that our system can generate naturalistic spoken dialogue while maintaining high semantic coherence.

Pseudo-labeling based semi-supervised learning can mitigate the performance degradation resulting from the absence of labeled data in the target domain. In pseudo-labeling, the quality of pseudo-labels is crucial for the final performance. However, most works overlook the potential benefits of using decoder for pseudo-labels within the the mainstream hybrid Connectionist Temporal Classification (CTC) and attention (CTC/attention) based ASR architecture. Therefore, we propose Hybrid Pseudo-Labeling (HPL) to improve the quality of pseudo-labels during online decoding. HPL introduces a second-stage decoding using the decoder to alleviate substitution errors arising from the conditional independence assumption inherent in CTC for error correction. Furthermore, we propose Hybrid Selection to optimally combine results of encoder and decoder. Additionally, we introduce Speed Perturbation Enhancement (SPE) to further enhance the quality of pseudo-labels via speed perturbation. Experiments demonstrate that HPL achieves state-of-the-art performance compared to other mainstream pseudo-labeling methods.

Conversational short-phrase speaker diarization focuses on diarizing the phrases that are short in duration. Nonetheless, conventional speaker diarization systems fail to give enough importance to conversational short phrases. This letter proposed a novel speaker diarization system to address this issue. Firstly, we employ an RNN-T model for joint speech recognition and speaker change detection. The speech recognition results can be utilized directly in downstream tasks while the speaker change points serve as guidance for the following steps. Secondly, we introduce self-adjusting speech segmentation, which dynamically adjusts segment lengths based on the temporal distribution of speaker change points. Thirdly, we introduce self-adjusting embedding extraction, which employs speaker encoders trained under different speech duration conditions by projecting them to the same embedding space. Our method achieves a major reduction of Diarization Error Rate (DER) and Conversational Diarization Error Rate (CDER) on the MagicData-RAMC and Mixer 6 datasets.

For efficient transmission of speech signals, speech compression methodologies have attracted significant research attention for decades and are widely used in automatic speech recognition (ASR) services. However, most speech codecs are perception-oriented, leaving redundant information and introducing distortion, which harms ASR systems. Recently, the emergence of neural network-based models has significantly advanced the progress of ASR systems and speech coding, laying the foundation for building a speech compression method specially optimized for ASR systems. In this letter, we propose an ASR-oriented Speech Quantization (ASQ) method to reduce communication costs for speech recognition systems. In the proposed method, a speech quantization model first converts the speech into low bit rate tokens. Then the tokens are transmitted to the server and recognized by a quantized speech recognition model. The two models could be jointly trained in the end-to-end (E2E) style. To mitigate the performance degradation introduced by the quantization components, we design an entropy-guided 3-stage training method that encourages the model to fully utilize the token space and promote recognition accuracy. Experiment results on the LibriSpeech corpus show that compared to an existing non-quantized ASR model with a 256 kbps transmission bit rate, the proposed method can achieve a transmission bit rate of 0.6 kbps without any influence on word error rate (WER). It also significantly surpasses the 2-step pipeline that first performs speech codec and then recognizes with a several times lower bit rate.

In clustering-based speaker diarization systems, the embedding clusters for distinctive speakers exhibit wide variability in size and density, posing difficulty for clustering accuracy. In spite of this, with the assistance of the overall distance relationships among speaker embeddings, most of the embeddings can be grouped to the correct cluster by sophisticated offline clustering algorithms. However, in online scenarios, such a complete distance relationships of the embeddings can not be obtained due to the incremental arrival of embeddings. Consequently, determining the number of clusters and then correctly grouping the embeddings become challenging in an online fashion. Furthermore, errors would accumulate quickly over time if the online clustering algorithm assigns the embeddings into clusters erroneously in the beginning. To address these problems, we designed a novel framework for online clustering. To reduce the high variability of speaker embeddings, we proposed the clustering guided embedding extractor training (CGEET) algorithm to encourage similarity between the size of the embedding space for different speakers in attempt to simplify the distance relationships of embeddings. The CGEET algorithm can grasp the distance information of the entire speaker embedding space and provide it to the online clustering algorithm. With this preliminary information, the distance thresholds guided online clustering (DTGOC) algorithm then processes incoming embeddings using a divide-and-conquer approach. It first handles the embeddings with explicit distance relationships and then searches for possible path combination they have with remaining embeddings in an online fashion. Moreover, in order to utilize the distance relationships of embeddings that are far apart in time, an online re-clustering strategy is incorporated in our DTGOC algorithm, which can alleviate error accumulation during online clustering. By implementing the above innovations, our proposed online clustering system achieves 14.00% DER with collar 0.25 at 2.5 s latency on the AISHELL-4, while the DER of the offline agglomerative hierarchical clustering system is 14.54%.

The cross-domain performance of automatic speech recognition (ASR) could be severely hampered due to the mismatch between training and testing distributions. Since the target domain usually lacks labeled data, and domain shifts exist at acoustic and linguistic levels, it is challenging to perform unsupervised domain adaptation (UDA) for ASR. Previous work has shown that self-supervised learning (SSL) or pseudo-labeling (PL) is effective in UDA by exploiting the self-supervisions of unlabeled data. However, these self-supervisions also face performance degradation in mismatched domain distributions, which previous work fails to address. This work presents a systematic UDA framework to fully utilize the unlabeled data with self-supervision in the pre-training and fine-tuning paradigm. On the one hand, we apply continued pre-training and data replay techniques to mitigate the domain mismatch of the SSL pre-trained model. On the other hand, we propose a domain-adaptive fine-tuning approach based on the PL technique with three unique modifications: Firstly, we design a dual-branch PL method to decrease the sensitivity to the erroneous pseudo-labels; Secondly, we devise an uncertainty-aware confidence filtering strategy to improve pseudo-label correctness; Thirdly, we introduce a two-step PL approach to incorporate target domain linguistic knowledge, thus generating more accurate target domain pseudo-labels. Experimental results on various cross-domain scenarios demonstrate that the proposed approach effectively boosts the cross-domain performance and significantly outperforms previous approaches.

When labeled data is insufficient, semi-supervised learning with the pseudo-labeling technique can significantly improve the performance of automatic speech recognition. However, pseudo-labels are often noisy, containing numerous incorrect tokens. Taking noisy labels as ground-truth in the loss function results in suboptimal performance. Previous works attempted to mitigate this issue by either filtering out the nosiest pseudo-labels or improving the overall quality of pseudo-labels. While these methods are effective to some extent, it is unrealistic to entirely eliminate incorrect tokens in pseudo-labels. In this work, we propose a novel framework named alternative pseudo-labeling to tackle the issue of noisy pseudo-labels from the perspective of the training objective. The framework comprises several components. Firstly, a generalized CTC loss function is introduced to handle noisy pseudo-labels by accepting alternative tokens in the positions of incorrect tokens. Applying this loss function in pseudo-labeling requires detecting incorrect tokens in the predicted pseudo-labels. In this work, we adopt a confidence-based error detection method that identifies the incorrect tokens by comparing their confidence scores with a given threshold, thus necessitating the confidence score to be discriminative. Hence, the second proposed technique is the contrastive CTC loss function that widens the confidence gap between the correctly and incorrectly predicted tokens, thereby improving the error detection ability. Additionally, obtaining satisfactory performance with confidence-based error detection typically requires extensive threshold tuning. Instead, we propose an automatic thresholding method that uses labeled data as a proxy for determining the threshold, thus saving the pain of manual tuning. Experiments demonstrate that alternative pseudo-labeling outperforms existing pseudo-labeling approaches on datasets in various domains and languages.

Recently, end-to-end automatic speech recognition models based on connectionist temporal classification (CTC) have achieved impressive results, especially when fine-tuned from wav2vec2.0 models. Due to the conditional independence assumption, CTC-based models are always weaker than attention-based encoder-decoder models and require the assistance of external language models (LMs). To solve this issue, we propose two knowledge transferring methods that leverage pre-trained LMs, such as BERT and GPT2, to improve CTC-based models. The first method is based on representation learning, in which the CTC-based models use the representation produced by BERT as an auxiliary learning target. The second method is based on joint classification learning, which combines GPT2 for text modeling with a hybrid CTC/attention architecture. Experiment on AISHELL-1 corpus yields a character error rate (CER) of 4.2% on the test set. When compared to the vanilla CTC-based models fine-tuned from the wav2vec2.0 models, our knowledge transferring method reduces CER by 16.1% relatively without external LMs.

While Transformers have achieved promising results in end-to-end (E2E) automatic speech recognition (ASR), their autoregressive (AR) structure becomes a bottleneck for speeding up the decoding process. For real-world deployment, ASR systems are desired to be highly accurate while achieving fast inference. Non-autoregressive (NAR) models have become a popular alternative due to their fast inference speed, but they still fall behind AR systems in recognition accuracy. To fulfill the two demands, in this paper, we propose a NAR CTC/attention model utilizing both pre-trained acoustic and language models: wav2vec2.0 and BERT. To bridge the modality gap between speech and text representations obtained from the pre-trained models, we design a novel modality conversion mechanism, which is more suitable for logographic languages. During inference, we employ a CTC branch to generate a target length, which enables the BERT to predict tokens in parallel. We also design a cache-based CTC/attention joint decoding method to improve the recognition accuracy while keeping the decoding speed fast. Experimental results show that the proposed NAR model greatly outperforms our strong wav2vec2.0 CTC baseline (15.1% relative CER reduction on AISHELL-1). The proposed NAR model significantly surpasses previous NAR systems on the AISHELL-1 benchmark and shows a potential for English tasks.

For online speaker diarization, samples arrive incrementally, and the overall distribution of the samples is invisible.Moreover, in most existing clustering-based methods, the training objective of the embedding extractor is not designed specially for clustering.To improve online speaker diarization performance, we propose a unified online clustering framework, which provides an interactive manner between embedding extractors and clustering algorithms.Specifically, the framework consists of two highly coupled parts: clustering-guided recurrent training (CGRT) and truncated beam searching clustering (TBSC).The CGRT introduces the clustering algorithm into the training process of embedding extractors, which could provide not only cluster-aware information for the embedding extractor, but also crucial parameters for the clustering process afterward.And with these parameters, which contain preliminary information of the metric space, the TBSC penalizes the probability score of each cluster, in order to output more accurate clustering results in online fashion with low latency.With the above innovations, our proposed online clustering system achieves 14.48% DER with collar 0.25 at 2.5s latency on the AISHELL-4, while the DER of the offline agglomerative hierarchical clustering is 14.57%.

There is often a trade-off between performance and latency in streaming automatic speech recognition (ASR).Traditional methods such as look-ahead and chunk-based methods, usually require information from future frames to advance recognition accuracy, which incurs inevitable latency even if the computation is fast enough.A causal model that computes without any future frames can avoid this latency, but its performance is significantly worse than traditional methods.In this paper, we propose corresponding revision strategies to improve the causal model.Firstly, we introduce a real-time encoder states revision strategy to modify previous states.Encoder forward computation starts once the data is received and revises the previous encoder states after several frames, which is no need to wait for any right context.Furthermore, a CTC spike position alignment decoding algorithm is designed to reduce time costs brought by the proposed revision strategy.Experiments are all conducted on Librispeech datasets.Fine-tuning on the CTC-based wav2vec2.0model, our best method can achieve 3.7/9.2WERs on test-clean/other sets and brings 45% relative improvement for causal models, which is also competitive with the chunkbased methods and the knowledge distillation methods.

This paper introduces a high-quality rich annotated Mandarin conversational (RAMC) speech dataset called MagicData-RAMC.The MagicData-RAMC corpus contains 180 hours of conversational speech data recorded from native speakers of Mandarin Chinese over mobile phones with a sampling rate of 16 kHz.The dialogs in MagicData-RAMC are classified into 15 diversified domains and tagged with topic labels, ranging from science and technology to ordinary life.Accurate transcription and precise speaker voice activity timestamps are manually labeled for each sample.Speakers' detailed information is also provided.As a Mandarin speech dataset designed for dialog scenarios with high quality and rich annotations, MagicData-RAMC enriches the data diversity in the Mandarin speech community and allows extensive research on a series of speechrelated tasks, including automatic speech recognition, speaker diarization, topic detection, keyword search, text-to-speech, etc.We also conduct several relevant tasks and provide experimental results to help evaluate the dataset.

Self-supervised pre-training could effectively improve the performance of low-resource automatic speech recognition (ASR).However, existing self-supervised pre-training are taskagnostic, i.e., could be applied to various downstream tasks.Although it enlarges the scope of its application, the capacity of the pre-trained model is not fully utilized for the ASR task, and the learned representations may not be optimal for ASR.In this work, in order to build a better pre-trained model for low-resource ASR, we propose a pre-training approach called wav2vec-S, where we use task-specific semi-supervised pretraining to refine the self-supervised pre-trained model for the ASR task thus more effectively utilize the capacity of the pretrained model to generate task-specific representations for ASR.Experiments show that compared to wav2vec 2.0, wav2vec-S only requires a marginal increment of pre-training time but could significantly improve ASR performance on in-domain, cross-domain and cross-lingual datasets.Average relative WER reductions are 24.5% and 6.6% for 1h and 10h fine-tuning, respectively.Furthermore, we show that semi-supervised pretraining could close the representation gap between the selfsupervised pre-trained model and the corresponding fine-tuned model through canonical correlation analysis.

Automatic speech recognition (ASR) with federated learning (FL) makes it possible to leverage data from multiple clients without compromising privacy. The quality of FL-based ASR could be measured by recognition performance, communication and computation costs. When data among different clients are not independently and identically distributed (non-IID), the performance could degrade significantly. In this work, we tackle the non-IID issue in FL-based ASR with personalized FL, which learns personalized models for each client. Concretely, we propose two types of personalized FL approaches for ASR. Firstly, we adapt the personalization layer based FL for ASR, which keeps some layers locally to learn personalization models. Secondly, to reduce the communication and computation costs, we propose decoupled federated learning (DecoupleFL). On one hand, DecoupleFL moves the computation burden to the server, thus decreasing the computation on clients. On the other hand, DecoupleFL communicates secure high-level features instead of model parameters, thus reducing communication cost when models are large. Experiments demonstrate two proposed personalized FL-based ASR approaches could reduce WER by 2.3% - 3.4% compared with FedAvg. Among them, DecoupleFL has only 11.4% communication and 75% computation cost compared with FedAvg, which is also significantly less than the personalization layer based FL.

Text-to-speech alignment, also known as time alignment, is essential for automatic speech recognition (ASR) systems used for speech retrieval tasks, such as keyword search and speech segment extraction. Previous works have used the Gaussian mixture model-hidden Markov model (GMM-HMM) forced alignment to improve the alignment performance. However, when used with end-to-end (E2E) ASR, GMM-HMM forced alignment causes extra reliance on expertise such as pronunciation lexica. It also increases the system complexity because GMM-HMMs are very dissimilar to E2E models. To tackle these two problems, we propose an E2E-ASR-based iteratively-trained timestamp estimator (ITSE), which performs alignment between token-level transcription and speech. We train ITSE first with coarse initial alignment targets generated using connectionist temporal classification (CTC) posteriors. During training, we iteratively perform realignment to update the targets. We attribute the effectiveness of the iterative training to ITSE’s two vital features. First, ITSE performs alignment using similarities between token and speech embeddings instead of frame-wise token classification posteriors. Second, ITSE uses speech embeddings that are aware of left context rather than global context. ITSE significantly outperforms CTC-based baselines in word alignment accuracy and is comparable to a GMM-HMM forced aligner. In short, ITSE is an accurate, lightweight text-to-speech alignment module implemented without expertise such as pronunciation lexica.

Even though attention-based end-to-end (E2E) automatic speech recognition (ASR) models have been yielding state-of-the-art recognition accuracy, they still fall behind many of the ASR models deployed in the industry in some crucial functionalities such as online processing and precise timestamps generating. This weakness prevents attention-based E2E ASR models from being applied in several essential speech tasks, such as online speech recognition and keyword searching (KWS). In this paper, we describe our proposed unified attention-based E2E ASR and KWS architecture–ETEH, which supports, in one model, both online and offline ASR decoding modes, thus allowing for precise and reliable KWS. “ETE” stands for attention-based E2E modeling, whereas “H” represents the hybrid gaussian mixture model and hidden Markov model (GMM-HMM). As a combination of both, ETEH is an attention-based E2E ASR architecture which utilizes the frame-wise time alignment (FTA) generated by GMM-HMM ASR models. This FTA is used to better the model in two ways: first, it helps the monotonic attentions of ETEH models to capture more accurate word time stamps, thus resulting in lower latency for online decoding; second, it helps ETEH models to provide accurate and reliable KWS results. Furthermore, we are able to combine both offline and online modes in one ETEH model and establish a concise system by adopt the universal training strategy. ETEH is functional and unique, and to the best of our knowledge, we can hardly find a comparable single attention-based E2E ASR system as the baseline. To evaluate ASR accuracy and latency for ETEH, we use our previously proposed monotonic truncated attention (MTA) based online CTC/attention (OCA) ASR models as baselines. Experimental results show that ETEH ASR models gain significant improvement in ASR latency compared to the baseline. To evaluate KWS performance, we compare ETEH models with CTC-based KWS models. Results demonstrate that our ETEH models achieve significantly better KWS performance compared to the CTC baselines.

Recently, we have witnessed excellent improvement of end-to-end (E2E) automatic speech recognition (ASR). However, how to tackle the long-tailed data distribution problem while maintaining E2E ASR models' performance for high-frequency tokens is still challenging. To solve this challenge, we propose a novel decoupled ASR learning method for the sequence-to-sequence ASR architecture in this paper. Our method decouples the learning procedure of this model into two stages: representation learning and classification learning. In the representation learning stage, we use the encoder output of a pretrained language model as one of the ASR model’s learning targets, and propose threshold log cosine embedding loss (TLCE-loss) as the objective function. A frequency-mask cross-entropy loss (FMCE-loss) is also designed as an auxiliary loss. In the classification learning stage, we find that introducing a temperature into softmax function helps reduce the influence of negative samples on tail classes, thus mitigating the biased learning process for the classifier. Furthermore, we propose a weighted softmax (w-softmax) to adjust ASR posterior probabilities according to the token appearing frequency during inference. Additionally, we introduce tail word/character error rate (TWER / TCER) and head word/character error rate (HWER / HCER) that respectively evaluate the ASR accuracy for tail and head words/characters. Experimental results on the Switchboard and HKUST corpora show that our proposed method greatly outperforms the baseline, especially in TWER / TCER reduction. To the best of our knowledge, this is the first work to use a decoupled ASR learning method to alleviate the long-tailed problem in sequence-to-sequence ASR.

End-to-end (E2E) models, including the attention-based encoder-decoder (AED) models, have achieved promising performance on the automatic speech recognition (ASR) task. However, the supervised training process of the E2E model needs a large amount of speech-text paired data. In contrast, self-supervised pre-training can pre-train the model on the unlabeled data and then fine-tune it on the limited labeled data to realize better performance. Most of the previous self-supervised pre-training methods focus on learning hidden representations from speech but ignore how to utilize the unpaired text. As a result, previous works often pre-train an acoustic encoder and then fine-tune it as a classification based ASR model, such as Connectionist Temporal Classification (CTC) based model, rather than an AED model. In this paper, we propose a self-supervised pre-training method for the AED model (SP-AED). The SP-AED method contains acoustic pre-training for the encoder, linguistic pre-training for the decoder, and an adaptive combination fine-tuning for the whole system. We first design a linguistic pre-training method for decoder by utilizing the text-only data. The decoder will be pre-trained as a noise-condition language model to learn the prior distribution of the text. Then, we pre-train the AED encoder with the wav2vec2.0 method with some modifications. Finally, we combine the pre-trained encoder and decoder and fine-tune them on the limited labeled data. We design an adaptive combination method during fine-tuning by modifying the decoder’s input and output to prevent catastrophic forgetting. Experiments prove that compared with the random initialized models, the SP-AED pre-trained models can realize up to 17% relative improvement. And with similar model size or computational cost, we can get comparable results to other classification-based models on both English and Chinese corpus.

Far-field automatic speech recognition (ASR) is a challenging task due to the background noise and reverberation. To address this issue, we introduce a novel end-to-end multi-channel far-field ASR architecture. First, we use a complex-valued CNN based architecture designed for speech tasks as a neural beamformer. Second, we propose an auxiliary mod-ule called absolute position regression module (APRM) with a position prediction loss to help the neural beamformer be better aware of the corresponding frequencies of each input time-frequency (T-F) bin. Third, inspired by the short-term stationarity of human speech, we propose an approach called the Inter-Frame Similarity Difference (IFSD) method to au-tomatically select useful channels as the inputs of the ASR backend from the outputs of the neural beamformer. We also implement a complex-valued attention module for the output channels of the neural beamformer to utilize each other's in-formation, thereby preventing the final outputs from information loss. With the above innovations, our proposed model achieves 9.7% and 11.1% relative WER reductions over a DNN-MVDR baseline on the CHiME4 dataset and a dataset simulated using the Librispeech corpus.

History utterances contain rich contextual information; however, better extracting information from the history utterances and using it to improve the language model (LM) is still challenging. In this paper, we propose the history utterance embedding Transformer LM (HTLM), which includes an embedding generation network for extracting contextual information contained in the history utterances and a main Transformer LM for current prediction. In addition, the two-stage attention (TSA) is proposed to encode richer contextual information into the embedding of history utterances (h-emb) while supporting GPU parallel training. Furthermore, we combine the extracted h-emb and embedding of current utterance (c-emb) through the dot-product attention and a fusion method for HTLM's current prediction. Experiments are conducted on the HKUST dataset and achieve a 23.4% character error rate (CER) on the test set. Compared with the baseline, the proposed method yields 12.86 absolute perplexity reduction and 0.8% absolute CER reduction.

This paper presents a method to pre-train transformer-based encoder-decoder automatic speech recognition (ASR) models using sufficient target-domain text. During pre-training, we train the transformer decoder as a conditional language model with empty or artifical states, rather than the real encoder states. By this pre-training strategy, the decoder can learn how to generate grammatical text sequence before learning how to generate correct transcriptions. Contrast to other methods which utilize text only data to improve the ASR performance, our method does not change the network architecture of the ASR model or introduce extra component like text-to-speech (TTS) or text-to-encoder (TTE). Experimental results on LibriSpeech corpus show that the proposed method can relatively reduce the word error rate over 10%, using 960 hours transcriptions.

Attention-based end-to-end (E2E) automatic speech recognition (ASR) architectures are now the state-of-the-art in terms of recognition performance. However, despite their effectiveness, they have not been widely applied in keyword search (KWS) tasks yet. In this paper, we propose the Att-E2E-KWS architecture, an attention-based E2E ASR framework for KWS that can afford accurate and reliable keyword retrieval results. First, we design a basic framework to carry out KWS based on attention-based E2E ASR. We adopt the connectionist temporal classification and attention (CTC/Att) joint E2E ASR architecture and exploit the spike posterior property of CTC to provide the keywords time stamps. Second, we introduce the frame-synchronous phonemes modeling and use the dynamic programming (DP) algorithm to provide alignments between E2E grapheme outputs and phoneme outputs. We call this alignment procedure dynamic time alignment (DTA), which can provide the proposed Att-E2E-KWS system with more accurate time stamps and reliable confidence scores. Third, we use the Transformer, a self-attention-based encoder-decoder neural network, in place of conventional recurrent neural networks in order to yield more parallelizable models and increased training speed. We conduct comprehensive experiments on English and Mandarin Chinese. To the best of our knowledge, this is the first practical Att-E2E-KWS framework, and experimental results on Switchboard and HKUST corpora show that our proposed Att-E2E-KWS systems significantly outperform the CTC E2E ASR based KWS baselines.

Recently, Transformer has gained success in automatic speech recognition (ASR) field. However, it is challenging to deploy a Transformer-based end-to-end (E2E) model for online speech recognition. In this paper, we propose the Transformer-based online CTC/attention E2E ASR architecture, which contains the chunk self-attention encoder (chunk-SAE) and the monotonic truncated attention (MTA) based self-attention decoder (SAD). Firstly, the chunk-SAE splits the speech into isolated chunks. To reduce the computational cost and improve the performance, we propose the state reuse chunk-SAE. Sencondly, the MTA based SAD truncates the speech features monotonically and performs attention on the truncated features. To support the online recognition, we integrate the state reuse chunk-SAE and the MTA based SAD into online CTC/attention architecture. We evaluate the proposed online models on the HKUST Mandarin ASR benchmark and achieve a 23.66% character error rate (CER) with a 320 ms latency. Our online model yields as little as 0.19% absolute CER degradation compared with the offline baseline, and achieves significant improvement over our prior work on Long Short-Term Memory (LSTM) based online E2E models.

Recently, there has been increasing progress in end-to-end automatic speech recognition (ASR) architecture, which transcribes speech to text without any pre-trained alignments. One popular end-to-end approach is the hybrid Connectionist Temporal Classification (CTC) and attention (CTC/attention) based ASR architecture, which utilizes the advantages of both CTC and attention. The hybrid CTC/attention ASR systems exhibit performance comparable to that of the conventional deep neural network (DNN)/ hidden Markov model (HMM) ASR systems. However, how to deploy hybrid CTC/attention systems for online speech recognition is still a non-trivial problem. This article describes our proposed online hybrid CTC/attention end-to-end ASR architecture, which replaces all the offline components of conventional CTC/attention ASR architecture with their corresponding streaming components. Firstly, we propose stable monotonic chunk-wise attention (sMoChA) to stream the conventional global attention, and further propose monotonic truncated attention (MTA) to simplify sMoChA and solve the training-and-decoding mismatch problem of sMoChA. Secondly, we propose truncated CTC (T-CTC) prefix score to stream CTC prefix score calculation. Thirdly, we design dynamic waiting joint decoding (DWJD) algorithm to dynamically collect the predictions of CTC and attention in an online manner. Finally, we use latency-controlled bidirectional long short-term memory (LC-BLSTM) to stream the widely-used offline bidirectional encoder network. Experiments with LibriSpeech English and HKUST Mandarin tasks demonstrate that, compared with the offline CTC/attention model, our proposed online CTC/attention model improves the real time factor in human-computer interaction services and maintains its performance with moderate degradation. To the best of our knowledge, this is the first work to provide the full-stack online solution for CTC/attention end-to-end ASR architecture.