基于自适应心理声学模型的智能语音识别系统

Intelligent speech recognition system based on self-adaption psychoacoustic model

针对包含环境噪声和信道失真等噪声的语音处理问题,提出了一种基于自适应心理声学模型的智能语音识别系统,并建立了听觉模型.该模型将心理声学和耳声发射(OAE)合并到了自动语音识别(ASR)系统中,利用AURORA2数据库分别在清洁训练条件和多训练条件下进行试验.结果表明,所提出的特征提取方法可以显著提高词识别率,优于梅尔频率倒谱系数(MFCC)、前向掩蔽(FM)、侧向抑制(LI)和倒谱平均值及方差归一化(CMVN)算法,能够有效地提高智能语音识别系统的性能.

Aiming at such noise speech processing problems as environmental noise and channel distortion,an intelligent speech recognition system based on adaptive psychoacoustic system was proposed,and an auditory model was established. In the proposed model,the psychoacoustics and otoacoustic emission（OAE） were integrated into an automatic speech recognition（ASR） system. With the AURORA2 database,the experiments were performed under both clean and multiple training conditions,respectively.The results showthat the proposed feature extraction method can significantly improve the word recognition rate,is superior to those of Mel-frequency cepstral coefficients（MFCCs）,forward masking（FM）,lateral inhibition（LI） and cepstral mean variance normalization（CMVN） algorithms,and can effectively enhance the performance of intelligent speech recognition system.