Education, Science, Technology, Innovation and Life
Open Access
Sign In
  1. Home
  2. Journals
  3. Advances in Computer, Signals and Systems
  4. Vol 7, Issue 3, 2023

Tibetan Speech Recognition Based on Wav2vec Feature

Download as PDF

DOI: 10.23977/acss.2023.070302 | Downloads: 15 | Views: 457

Author(s)

Zixi Yan 1, Guanyu Li 1, Senyan Li 1

Affiliation(s)

1 Northwest Minzu University, Lanzhou, Gansu, 730000, China

Corresponding Author

Guanyu Li

ABSTRACT

Speech recognition tasks for small languages such as Tibetan language have been unable to achieve the same results as those for large languages. In this paper, the wav2vec2 model is introduced into the traditional model to extract features and improve the effect of Tibetan speech recognition. In this paper, the kaldi tool was used to construct a speech recognition system for Tibetan language, and the wav2vec2 model was used as the feature extractor to replace the traditional mfcc features. The improvement effect of the front-end model and traditional speech recognition model on speech recognition of minority languages such as Tibetan wa comparatively anayzed, and the effectiveness of wav2vec2 model in non-pre-trained languages was verified. Finally, the recognition efficiency of the proposed model on per and wer was increased by 2.92% and 5.24% respectively as compared with the baseline system.

KEYWORDS

Tibetan, wav2vec2, speech recognition, kaldi, low resource

CITE THIS PAPER

Zixi Yan, Guanyu Li, Senyan Li. Tibetan Speech Recognition Based on Wav2vec Feature. Advances in Computer, Signals and Systems (2023) Vol. 7: 9-12. DOI: http://dx.doi.org/10.23977/acss.2023.070302.

REFERENCES

[1] Liu Jingping and Dexi Jiacuo, Design of ando-tibetan consonant recognition, Research on Ethnic Language Information Technology -- The 11th National Ethnic Language Information Symposium. 2007: p. 11-15.
[2] Müller and Meinard, Information retrieval for music and motion. 2007: Springer Berlin Heidelberg. 69-84.
[3] Rebello Sinda and Y.H. Y, An integrated approach for system functional reliability assessment using Dynamic Bayesian Network and Hidden Markov Model. Reliability Engineering & System Safety, 2018: p. 124-135.
[4] Pei Chunbao, Research on Tibetan Speech Recognition Technology based on Standard Lhasa language, 2009, Tibet University: Lhasa.
[5] Li Guanyu and Meng meng, Study on Acoustic Model for Continuous Large Vocabulary Speech Recognition of Tibetan Lhasa Dialect. Computer Engineering, 2012: p.189-191.
[6] Baevski A., et al., wav2vec 2.0 A Framework for Self-Supervised. 2020.
[7] Baevski A., S. Schneider and M. Auli, vq-wav2vec- Self-Supervised Learning of Discrete Speech Representations. 2020.
[8] Conneau A., et al., {Unsupervised Cross-lingual Representation Learning for Speech Recognition}. arXiv e-prints, 2020: p. arXiv:2006.13979.
[9] Yin Chen and Wu Min, A review of N-gram model. Computer Systems & Applications, 2018. 27(10): p.33-38.
[10] Xu Q., A. Baevski and M. Auli, Simple and Effective Zero-shot Cross-lingual Phoneme Recognition. arXiv e-prints, 2021: p. arXiv:2109.11680.

Subscription
E-Mail Alert
Most popular papers
Call for Special Issue
Publication Ethics & OA Statement
Plagiarism Policy
Contact Us
Downloads: 13338
Visits: 257235

Sponsors, Associates, and Links

All published work is licensed under a Creative Commons Attribution 4.0 International License.

Copyright © 2016 - 2031 Clausius Scientific Press Inc. All Rights Reserved.