A Rejection Model Based on Multi-Layer Perceptrons for Mandarin Digit Recognition

High performance Mandarin digit recognition(MDR)is much more difficult to achieve than its English counterpart,especially on inexpensive hardware implementation.In this paper,a new ,Multi-Layer Perceptrons(MLP)based postprocessor,an a posteriori probability estimator is presented and used for the rejection model of the speaker independent Mandarin digit recognition system based on hidden Markov model(HMM).Poor utterances,which are recognized by HMMs but have low a posteriori probability,will be rejected.After rejecting about 4.9% of the tested utteraces,the MLP rejection model can boost the digit recognition accuracy from 97.1%to 99.6%,The performance is better than those rejection models based on linear discrimiantion,likelihood ratio or anti-digit.

A rejection model based on multi-layer perceptrons for Mandarin digit recognition

High performance Mandarin digit recognition (MDR) is much more difficult to achieve than its English counterpart, especially on inexpensive hardware implementation. In this paper, a new Multi-Layer Perceptrons (MLP) based postprocessor, ana posteriori probability estimator, is presented and used for the rejection model of the speaker independent Mandarin digit recognition system based on hidden Markov model (HMM). Poor utterances, which are recognized by HMMs but have lowa posteriori probability, will be rejected. After rejecting about 4.9% of the tested utterances, the MLP rejection model can boost the digit recognition accuracy from 97.1% to 99.6%. The performance is better than those rejection models based on linear discrimination, likelihood ratio or anti-digit. This project is supported by the National Natural Science Fundation of China (Grant No.69975007) and the National “863” High-Tech Programme of China (No.863-306-ZD13-04-6), Open Funds of National Laboratory of Pattern Recognition, and Intel Architecture Development Co., Ltd. ZHONG Lin received his B.S. and M.S. degrees in circuit and system from Tsinghua University, Beijing, China, in 1998 and 2000, respectively. Now he is a Ph.D. candidate in the Electronic Engineering Department, Princeton University, US. LIU Jia received his B.S., M.S., and Ph.D. degrees in communication and electronic systems from Tsinghua University, Beijing, China, in 1983, 1986 and 1990, respectively. In April 1990, he joined the Remote Sensing Satellite Ground Station, Chinese Academy of Sciences, and then he worked as a Royal Society visiting scientist at the Engineering Department, Cambridge University, UK during 1992–1994. He is now a professor in the Department of Electronic Engineering, Tsinghua University and an IEEE member. His current research focuses on speech recognition, speech synthesis, speech coding, speech ASIC design and multimedia communication. LIU Runsheng graduated from the Department of Radio and Electronics, Tsinghua University, Beijing, China, in 1958. Since 1958, he has been working at Tsinghua University and he is now a professor in the Department of Electronic Engineering, Tsinghua University, where he teaches and conducts researches on digital and analog circuits, IC design, electronic circuit CAD, signal processing and digital communication.