嵌入式盲人手机语音识别与控制系统设计.pdf

. 2009. 17( 10) Computer Measurement Control 1951 : 1671- 4598(2009) 10- 1951- 04 :TP39142 :A 王璟王旬, 滕召胜, 高云鹏, 王 一, 李聪聪 ( , 410082) : , ; AD1836A Blackifn533 DSP, ( C M M) (M FCC) , Baum- Welch (MCE) MM , 10 (0~ 9) 25 , : ; Baum- Welch ; ; Design of Embedded Speech Recognition and Control System on Mobile Phone for the Blind Wang Jingxun, Teng Zhaosheng, Gao Yunpeng, Wang Yi, Li Congcong (College of Electrical Information Engineering, unan University, Changsha 410082, China) Abstract: On behalf of the blind and weak- sighted individuals, an embedded speech recognition and control system of mobile phoneis designed based on the speaker- independent speech recognition technique. A Blackfin533 DSP is chosen as the CPU to process thespeech signals sampled by the audio codec AD1836A. Thecontinuous idden Markov Model (C MM) is adopted for recognition, and the Melfrequency cepstral coefficient (MFCC) asfeatureparameters. By using theBaum- Welch algorithm totrain themodels andtheminimum classification error (MCE) methodto im prove model parameters, this system realizes a complete voice control platform containing 10 numbers (0~ 9) and 25 voice commands for the blind mobile phone users. A method of hierarchical recognition for speech commands is designed to improve the overall recognition accuracy. Key ords: hidden markov M odel; Baum- Welch algorithm; minimum classification error; mobile phone for the blind 0 , (C MM), [2] PC , , , DSP ,