arXiv subjects

Chunyang Wu

Publications and source records attributed to Chunyang Wu.

25 records · Page 2Linked to original sources

arXiv2020

Emformer: Efficient Memory Transformer Based Acoustic Model For Low Latency Streaming Speech Recognition

This paper proposes an efficient memory transformer Emformer for low latency streaming speech recognition. In Emformer, the long-range history context is distilled into an augmented memory bank to reduce self-attention's computation complexity. A cache mechanism saves the computation for the key and value in self-attention for the left context. Emformer applies a parallelized block processing in training to support low latency models. We carry out experiments on benchmark LibriSpeech data. Under average latency of 960 ms, Emformer gets WER $2.50\%$ on test-clean and $5.62\%$ on test-other. Comparing with a strong baseline augmented memory transformer (AM-TRF), Emformer gets $4.6$ folds training speedup and $18\%$ relative real-time factor (RTF) reduction in decoding with relative WER reduction $17\%$ on test-clean and $9\%$ on test-other. For a low latency scenario with an average latency of 80 ms, Emformer achieves WER $3.01\%$ on test-clean and $7.09\%$ on test-other. Comparing with the LSTM baseline with the same latency and model size, Emformer gets relative WER reduction $9\%$ and $16\%$ on test-clean and test-other, respectively.

cs.SD↗

arXiv2020

Streaming Attention-Based Models with Augmented Memory for End-to-End Speech Recognition

Attention-based models have been gaining popularity recently for their strong performance demonstrated in fields such as machine translation and automatic speech recognition. One major challenge of attention-based models is the need of access to the full sequence and the quadratically growing computational cost concerning the sequence length. These characteristics pose challenges, especially for low-latency scenarios, where the system is often required to be streaming. In this paper, we build a compact and streaming speech recognition system on top of the end-to-end neural transducer architecture with attention-based modules augmented with convolution. The proposed system equips the end-to-end models with the streaming capability and reduces the large footprint from the streaming attention-based model using augmented memory. On the LibriSpeech dataset, our proposed system achieves word error rates 2.7% on test-clean and 5.8% on test-other, to our best knowledge the lowest among streaming approaches reported so far.

cs.CL↗

arXiv2020

Transformer in action: a comparative study of transformer-based acoustic models for large scale speech recognition applications

In this paper, we summarize the application of transformer and its streamable variant, Emformer based acoustic model for large scale speech recognition applications. We compare the transformer based acoustic models with their LSTM counterparts on industrial scale tasks. Specifically, we compare Emformer with latency-controlled BLSTM (LCBLSTM) on medium latency tasks and LSTM on low latency tasks. On a low latency voice assistant task, Emformer gets 24% to 26% relative word error rate reductions (WERRs). For medium latency scenarios, comparing with LCBLSTM with similar model size and latency, Emformer gets significant WERR across four languages in video captioning datasets with 2-3 times inference real-time factors reduction.

cs.CL↗

arXiv2020

Weak-Attention Suppression For Transformer Based Speech Recognition

Transformers, originally proposed for natural language processing (NLP) tasks, have recently achieved great success in automatic speech recognition (ASR). However, adjacent acoustic units (i.e., frames) are highly correlated, and long-distance dependencies between them are weak, unlike text units. It suggests that ASR will likely benefit from sparse and localized attention. In this paper, we propose Weak-Attention Suppression (WAS), a method that dynamically induces sparsity in attention probabilities. We demonstrate that WAS leads to consistent Word Error Rate (WER) improvement over strong transformer baselines. On the widely used LibriSpeech benchmark, our proposed method reduced WER by 10%$ on test-clean and 5% on test-other for streamable transformers, resulting in a new state-of-the-art among streaming models. Further analysis shows that WAS learns to suppress attention of non-critical and redundant continuous acoustic frames, and is more likely to suppress past frames rather than future ones. It indicates the importance of lookahead in attention-based ASR models.

eess.AS↗

arXiv2020

Streaming Transformer-based Acoustic Models Using Self-attention with Augmented Memory

Transformer-based acoustic modeling has achieved great suc-cess for both hybrid and sequence-to-sequence speech recogni-tion. However, it requires access to the full sequence, and thecomputational cost grows quadratically with respect to the in-put sequence length. These factors limit its adoption for stream-ing applications. In this work, we proposed a novel augmentedmemory self-attention, which attends on a short segment of theinput sequence and a bank of memories. The memory bankstores the embedding information for all the processed seg-ments. On the librispeech benchmark, our proposed methodoutperforms all the existing streamable transformer methods bya large margin and achieved over 15% relative error reduction,compared with the widely used LC-BLSTM baseline. Our find-ings are also confirmed on some large internal datasets.

eess.AS↗

arXiv2016

A machine learning method for the large-scale evaluation of urban visual environment

Given the size of modern cities in the urbanising age, it is beyond the perceptual capacity of most people to develop a good knowledge about the beauty and ugliness of the city at every street corner. Correspondingly, for planners, it is also difficult to accurately answer questions like 'where are the worst-looking places in the city that regeneration should give first consideration', or 'in the fast urbanising cities, how is the city appearance changing', etc. To address this issue, we here present a computer vision method for the large-scale and automatic evaluation of the urban visual environment, by leveraging state-of-the-art machine learning techniques and the wide-coverage street view images. From the various factors that are at work, we choose two key features, the visual quality of street facade and the continuity of street wall, as the starting point of this line of analysis. In order to test the validity of this method, we further compare the machine ratings with ratings collected on site from 752 passers-by on fifty-six locations. We show that the machine learning model can produce a good estimation of people's real visual experience, and it holds much potential for various tasks in terms of urban design evaluation, culture identification, etc.

cs.CV↗

arXiv2015

Controlled Synthesis of Organic/Inorganic van der Waals Solid for Tunable Light-matter Interactions

Van der Waals (vdW) solids, as a new type of artificial materials that consist of alternating layers bonded by weak interactions, have shed light on fascinating optoelectronic device concepts. As a result, a large variety of vdW devices have been engineered via layer-by-layer stacking of two-dimensional materials, although shadowed by the difficulties of fabrication. Alternatively, direct growth of vdW solids has proven as a scalable and swift way, highlighted by the successful synthesis of graphene/h-BN and transition metal dichalcogenides (TMDs) vertical heterostructures from controlled vapor deposition. Here, we realize high-quality organic and inorganic vdW solids, using methylammonium lead halide (CH3NH3PbI3) as the organic part (organic perovskite) and 2D inorganic monolayers as counterparts. By stacking on various 2D monolayers, the vdW solids behave dramatically different in light emission. Our studies demonstrate that h-BN monolayer is a great complement to organic perovskite for preserving its original optical properties. As a result, organic/h-BN vdW solid arrays are patterned for red light emitting. This work paves the way for designing unprecedented vdW solids with great potential for a wide spectrum of applications in optoelectronics.

physics.chem-ph↗