This paper tackles several challenges that arise when integrating Automatic Speech Recognition (ASR) and Machine Translation (MT) for real-time, on-device streaming speech translation. Although state-of-the-art ASR systems based on Recurrent Neural Network Transducers (RNN-T) can perform real-time transcription, achieving streaming translation in real-time remains a significant challenge. To address this issue, we propose a simultaneous translation approach that effectively balances translation quality and latency. We also investigate efficient integration of ASR and MT, leveraging linguistic cues generated by the ASR system to manage context and utilizing efficient beam-search pruning techniques such as time-out and forced finalization to maintain system's real-time factor. We apply our approach to an on-device bilingual conversational speech translation and demonstrate that our techniques outperform baselines in terms of latency and quality. Notably, our technique narrows the quality gap with non-streaming translation systems, paving the way for more accurate and efficient real-time speech translation.

We propose a first streaming accent conversion (AC) model that transforms non-native speech into a native-like accent while preserving speaker identity, prosody and improving pronunciation. Our approach enables stream processing by modifying a previous AC architecture with an Emformer encoder and an optimized inference mechanism. Additionally, we integrate a native text-to-speech (TTS) model to generate ideal ground-truth data for efficient training. Our streaming AC model achieves comparable performance to the top AC models while maintaining stable latency, making it the first AC system capable of streaming.

There has been increasing interest in unifying streaming and non-streaming automatic speech recognition (ASR) models to reduce development, training, and deployment costs. W e present a unified framework that trains a single end-to-end ASR model for both streaming and non-streaming applications, leveraging future context information. W e propose to use dynamic right-context through the chunked attention masking in the training of zipformer-based ASR models. W e demonstrate that using right-context is more effective in zipformer models compared to other conformer models due to its multi-scale nature. W e analyze the effect of varying the number of right-context frames on accuracy and latency of the streaming ASR models. W e use Librispeech and large in-house conversational datasets to train different versions of streaming and non-streaming models and evaluate them in a production grade server-client setup across diverse testsets of different domains. The proposed strategy reduces word error by relative 7.9% with a small degradation in user-perceived latency. By adding more right-context frames, we are able to achieve streaming performance close to that of non-streaming models. Our approach also allows flexible control of the latency-accuracy tradeoff according to customers requirements.

Simultaneous or streaming machine translation generates translation while reading the input stream. These systems face a quality/latency trade-off, aiming to achieve high translation quality similar to non-streaming models with minimal latency. We propose an approach that efficiently manages this trade-off. By enhancing a pretrained non-streaming model, which was trained with a seq2seq mechanism and represents the upper bound in quality, we convert it into a streaming model by utilizing the alignment between source and target tokens. This alignment is used to learn a read/write decision boundary for reliable translation generation with minimal input. During training, the model learns the decision boundary through a read/write policy module, employing supervised learning on the alignment points (pseudo labels). The read/write policy module, a small binary classification unit, can control the quality/latency trade-off during inference. Experimental results show that our model outperforms several strong baselines and narrows the gap with the non-streaming baseline model.

Streaming automatic speech recognition (ASR) models are restricted from accessing future context, which results in worse performance compared to the non-streaming models. To improve the performance of streaming ASR, knowledge distillation (KD) from the non-streaming to streaming model has been studied, mainly focusing on aligning the output token probabilities. In this paper, we propose a layer-to-layer KD from the teacher encoder to the student encoder. To ensure that features are extracted using the same context, we insert auxiliary non-streaming branches to the student and perform KD from the non-streaming teacher layer to the non-streaming auxiliary layer. We design a special KD loss that leverages the autoregressive predictive coding (APC) mechanism to encourage the streaming model to predict unseen future contexts. Experimental results show that the proposed method can significantly reduce the word error rate compared to previous token probability distillation methods.

Transducer is one of the mainstream frameworks for streaming speech recognition. There is a performance gap between the streaming and non-streaming transducer models due to limited context. To reduce this gap, an effective way is to ensure that their hidden and output distributions are consistent, which can be achieved by hierarchical knowledge distillation. However, it is difficult to ensure the distribution consistency simultaneously because the learning of the output distribution depends on the hidden one. In this paper, we propose an adaptive two-stage knowledge distillation method consisting of hidden layer learning and output layer learning. In the former stage, we learn hidden representation with full context by applying mean square error loss function. In the latter stage, we design a power transformation based adaptive smoothness method to learn stable output distribution. It achieved 19\% relative reduction in word error rate, and a faster response for the first token compared with the original streaming model in LibriSpeech corpus.

Existing training criteria in automatic speech recognition(ASR) permit the model to freely explore more than one time alignments between the feature and label sequences. In this paper, we use entropy to measure a model's uncertainty, i.e. how it chooses to distribute the probability mass over the set of allowed alignments. Furthermore, we evaluate the effect of entropy regularization in encouraging the model to distribute the probability mass only on a smaller subset of allowed alignments. Experiments show that entropy regularization enables a much simpler decoding method without sacrificing word error rate, and provides better time alignment quality.

In models typically operate under more storage and computational constraints contrast, most non-streaming models run from the server with fewer - e.g., on embedded devices - than any server-side full-context constraints. Therefore, instead of developing equally sized encoders, models. Motivated by the recent progress in Omni-sparsity supernet it is preferable to jointly build a compact streaming model and a training, where multiple subnetworks are jointly optimized in one large non-streaming model for real-world ASR applications. We note single model, this work aims to jointly learn a compact sparse ondevice that even though a single encoder is shared for both modes, we can streaming ASR model, and a large dense server non-streaming substantially prune it into a featherweight, e.g., about 30M parameters model, in a single supernet. Next, we present that, performing supernet as a streaming model, and use the original copy as a performant nonstreaming training on both wav2vec 2.0 self-supervised learning and encoder. Given the recent progress made in neural network supervised ASR fine-tuning can not only substantially improve the pruning [18, 19, 20, 21], we can specify a target sparsity level during large non-streaming model as shown in prior works, and also be able model training, prune the model weights accordingly before inference, to improve the compact sparse streaming model.

The streaming automatic speech recognition (ASR) models are more popular and suitable for voice-based applications. However, non-streaming models provide better performance as they look at the entire audio context. To leverage the benefits of the non-streaming model in streaming applications like voice search, it is commonly used in second pass re-scoring mode. The candidate hypothesis generated using steaming models is re-scored using a non-streaming model. In this work, we evaluate the non-streaming attention-based end-to-end ASR models on the Flipkart voice search task in both standalone and re-scoring modes. These models are based on Listen-Attend-Spell (LAS) encoder-decoder architecture. We experiment with different encoder variations based on LSTM, Transformer, and Conformer. We compare the latency requirements of these models along with their performance. Overall we show that the Transformer model offers acceptable WER with the lowest latency requirements. We report a relative WER improvement of around 16% with the second pass LAS re-scoring with latency overhead under 5ms. We also highlight the importance of CNN front-end with Transformer architecture to achieve comparable word error rates (WER). Moreover, we observe that in the second pass re-scoring mode all the encoders provide similar benefits whereas the difference in performance is prominent in standalone text generation mode.