Speech Recognition Model Compression Algorithm Based on Bayesian Information Criterion

Abstract

Abstract:

Recognition rate of speech model will increase with the increase in the number of GMM components, the size of model will increase as well, when making the GMM

recognition training for HMM speech model, and it causes model bloated. However, it is unfit for mobile devices while using speech model for recognition to keep greater than

hundreds of megabytes in mobile. For this problem, a method for compress speech model based on BIC is presented. This method tries to keep recognition rate of speech model in

appropriate to the size of model. Experiments demonstrate that it’s applicable and available to achieve the final speech model specified size even ensure recognition rate of

speech model as much as possible.

Key words: speech recognition, model compress, BIC (bayesian information criterion)

ZOU Can, LI Bai-yan. Speech Recognition Model Compression Algorithm Based on Bayesian Information Criterion[J]. Computer and Modernization.

References

［1］
Jurafsky D, Martin.Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition(2nd ed)［M］. Prentice Hall,

2008.

［2］ Juang B H, Rabiner L R. Hidden Markov models for speech recognition［J］. Technometrics, 1991,33(3):251-272.
［3］ Xie Chen, Adam Eversole, Gang Li,et al. Pipelined Back-Propagation for Context-Dependent Deep Neural Networks［DB/OL］. http://research.microsoft.com/apps/pubs/?

id=173312, 2012-09-10.
［4］ Gideon Schwarz. Estimating the dimension of a model［J］. The Annals of Statistics, 1978,6(2):461-464.
［5］ Akaike H. A new look at the statistical identication model［J］. IEEE Transactions on Automatic Control, 1974,19(6):716-723.
［6］ Jin H, Kubala F, Schwartz R. Automatic speaker clustering［C］// Proceedings of the 1997 DARPA Speech Recognition Workshop. 1997:108-111.
［7］ Legetter C J, Woodland P C. Maximum likelihood linear regression for speaker adaptation of continuous density hidden Markov models［J］. Computer Speech and Language,

1995,9(2):171-185.
［8］ Geoffrey J McLachlan, Thriyambakam Krishnan. The EM Algorithm and Extensions(2nd ed)［M］. Wiley, 2008.
［9］ Lawrence Rabiner, Biing-Hwang Juang. Fundamentals of Speech Recognition［M］. USA: Prentice Hall, 1993.
［10］Akaike H. A new look at the statistical identification［J］. IEEE Transactions on Automatic Control, 1974,19(6):716-723.
［11］Schwarz G. A second-order approximation to optimal sampling regions［J］. The Annals of Mathematical Statistics, 1969,40(1):313-315.
［12］Schwarz G. A sequential student test［J］. The Annals of Mathematical Statistics, 1971,42(3):1003-1009.
［13］Tong H. Determination of the order of a Markov chain by Akaike’s information criterion［J］. Journal of Applied Probability, 1975,12(3):488-497.
［14］吴华,徐波,黄泰翼. 基于三音子模型的语料自动选择算法［J］. 软件学报, 2000,11(2):271-276.
［15］韩兆兵,贾磊,张树武,等. 连续语音识别中声学建模的组合聚类算法研究［J］. 中文信息学报, 2003,17(4):33-38.
［16］郝杰,李星. 汉语连续语音识别中经典HMM的实验评测［J］. 计算机工程与应用, 2001，37(13):1-4,101.

[1]	HUANG Yan-hui, LAN Hai, WEI Xian. Lightweight Vision Transformer Based on Separable Structured Transformations [J]. Computer and Modernization, 2022, 0(10): 75-81.
[2]	BAO Zhi-qiang, CHENG Ping, HUANG Qiong-dan, LYU Shao-qing. A Model Compression Algorithm of Convolutional Neural Network [J]. Computer and Modernization, 2021, 0(10): 107-111.
[3]	BAI Shi-lei, YIN Ke-xin, ZHU Jian-qi. Lightweight YOLOv3 Traffic Sign Detection Algorithm [J]. Computer and Modernization, 2020, 0(09): 83-88.
[4]	LIAO Xiao-dong, JIA Xiao-xia . Action Recognition Technology Based on Improved C3D Neural Network [J]. Computer and Modernization, 2019, 0(03): 32-.