Yiming Wang

Yiming Wang

Auto-generated profile, based on publicly available information

Yiming Wang

Center for Language and Speech Processing

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
18 March 2021 in IEEE Signal Processing Letters
Verified
Journal Article
Closed access
HLHang Lv
Hang Lv
DPDaniel Povey
Daniel Povey
MYMahsa Yarmohammadi
Mahsa Yarmohammadi
KLKe Li
Ke Li
YWYiming Wang
Yiming Wang
SKSanjeev Khudanpur
Sanjeev Khudanpur
We propose a novel lazy-evaluation token-group decoding algorithm with on-the-fly composition of weighted finite-state transducers (WFSTs) for large vocabulary continuous speech recognition. In the standard on-the-fly composition decoder, a base WFST and one or more incremental WFSTs are composed during decoding, and then token passing algorithm is employed to generate the lattice on the composed search space, resulting in substantial computation overhead. To improve speed, the proposed algorithm adopts 1) a token-group method, which groups tokens with the same state in the base WFST on each frame and limits the capacity of the group and 2) a lazy-evaluation method, which does not expand a token group and its source token groups until it processes a word label during decoding.Experiments show that the proposed decoder works notably up to 3 times faster than the standard on-the-fly composition decoder.

1 January 2025 in Electronics Letters
Verified
Journal Article
Open access
FLFangyi Li
Fangyi Li
HLHang Lv
Hang Lv
YWYiming Wang
Yiming Wang
LXLei Xie
Lei Xie
Neural language models (NLMs) have been shown to outperform n‐gram language models in automatic speech recognition (ASR) tasks. NLMs are usually used in the second‐pass lattice rescoring rather than the first‐pass decoding, since its encoded infinite history virtually cannot be compiled into static decoding graphs. However, the modeling power of NLMs is not fully leveraged due to the constraints imposed by the lattice, leading to accuracy loss. To improve this, on‐the‐fly composition decoders were proposed to utilize NLMs in first‐pass decoding with increased computational cost. In this paper, an asynchronous lazy‐evaluation token‐group decoder with exact lattice generation is proposed to reduce the computational cost of the on‐the‐fly composition decoder, achieving significant decoding speedup. More specifically, having a novel token‐group with a representative element data structure, the proposed decoder performs lazy‐evaluation which expands the tokens until a word boundary is reached. Furthermore, based on the score of the representative element in a token‐group, the decoder prunes unpromising tokens by an A* algorithm. The experiments show that the proposed decoder can accelerate the vanilla on‐the‐fly composition decoder by up to 6.9 times, and get paths with even better average likelihoods than lattice rescoring approaches.

27 July 2018 in arXiv
Verified
Preprint
Open access
PDF file
ZCZhehuai Chen
Zhehuai Chen
JLJustin Luitjens
Justin Luitjens
HXHainan Xu
Hainan Xu
YWYiming Wang
Yiming Wang
DPDaniel Povey
Daniel Povey
SKSanjeev Khudanpur
Sanjeev Khudanpur
We describe initial work on an extension of the Kaldi toolkit that supports weighted finite-state transducer (WFST) decoding on Graphics Processing Units (GPUs). We implement token recombination as an atomic GPU operation in order to fully parallelize the Viterbi beam search, and propose a dynamic load balancing strategy for more efficient token passing scheduling among GPU threads. We also redesign the exact lattice generation and lattice pruning algorithms for better utilization of the GPUs. Experiments on the Switchboard corpus show that the proposed method achieves identical 1-best results and lattice quality in recognition and confidence measure tasks, while running 3 to 15 times faster than the single process Kaldi decoder. The above results are reported on different GPU architectures. Additionally we obtain a 46-fold speedup with sequence parallelism and multi-process service (MPS) in GPU.

15 October 2019 in arXiv
Verified
Preprint
Open access
PDF file
YWYiming Wang
Yiming Wang
TCTongfei Chen
Tongfei Chen
HXHainan Xu
Hainan Xu
SDShuoyang Ding
Shuoyang Ding
HLHang Lv
Hang Lv
SKSanjeev Khudanpur
Sanjeev Khudanpur
We present Espresso, an open-source, modular, extensible end-to-end neural automatic speech recognition (ASR) toolkit based on the deep learning library PyTorch and the popular neural machine translation toolkit fairseq. Espresso supports distributed training across GPUs and computing nodes, and features various decoding approaches commonly employed in ASR, including look-ahead word-based language model fusion, for which a fast, parallelized decoder is implemented. Espresso achieves state-of-the-art ASR performance on the WSJ, LibriSpeech, and Switchboard data sets among other end-to-end systems without data augmentation, and is 4--11x faster for decoding than similar systems (e.g. ESPnet).

29 July 2020 in arXiv
Verified
Preprint
Open access
PDF file
YWYiming Wang
Yiming Wang
HLHang Lv
Hang Lv
DPDaniel Povey
Daniel Povey
Lei XieLei Xie
Lei Xie
SKSanjeev Khudanpur
Sanjeev Khudanpur
Always-on spoken language interfaces, e.g. personal digital assistants, rely on a wake word to start processing spoken input. We present novel methods to train a hybrid DNN/HMM wake word detection system from partially labeled training data, and to use it in on-line applications: (i) we remove the prerequisite of frame-level alignments in the LF-MMI training algorithm, permitting the use of un-transcribed training examples that are annotated only for the presence/absence of the wake word; (ii) we show that the classical keyword/filler model must be supplemented with an explicit non-speech (silence) model for good performance; (iii) we present an FST-based decoder to perform online detection. We evaluate our methods on two real data sets, showing 50%--90% reduction in false rejection rates at pre-specified false alarm rates over the best previously published figures, and re-validate them on a third (large) data set.

20 May 2020 in arXiv
Verified
Preprint
Open access
PDF file
YSYiwen Shao
Yiwen Shao
YWYiming Wang
Yiming Wang
DPDaniel Povey
Daniel Povey
SKSanjeev Khudanpur
Sanjeev Khudanpur
We present PyChain, a fully parallelized PyTorch implementation of end-to-end lattice-free maximum mutual information (LF-MMI) training for the so-called \emph{chain models} in the Kaldi automatic speech recognition (ASR) toolkit. Unlike other PyTorch and Kaldi based ASR toolkits, PyChain is designed to be as flexible and light-weight as possible so that it can be easily plugged into new ASR projects, or other existing PyTorch-based ASR tools, as exemplified respectively by a new project PyChain-example, and Espresso, an existing end-to-end ASR toolkit. PyChain's efficiency and flexibility is demonstrated through such novel features as full GPU training on numerator/denominator graphs, and support for unequal length sequences. Experiments on the WSJ dataset show that with simple neural networks and commonly used machine learning techniques, PyChain can achieve competitive results that are comparable to Kaldi and better than other end-to-end ASR systems.