基于HMM的非特定人汉语语音识别系统

HMM-based Chinese voice recognition system for nonspecific population

设计了基于隐马尔可夫模型(Hidden Markov Model,HMM)的非特定人汉语语音识别系统,主要由录音、训练和识别三大模块构成."录音模块"首先录制一段指定长度的语音信号,然后通过对语音信号的短时能量和过零率进行门限检测,标志出有效语音段并保存."训练模块"利用Baum-Welch算法计算语音样本的MFCC(Mel Frequency Cepstrum Coefficient)参数生成识别用的语音模板."识别模块"利用HMM识别算法比较语音信号和语音模板的相似概率,找到最大值输出,完成语音识别功能.最后,在MATLAB中实现了该语音识别系统.实验结果表明,系统的识别率为60%以上,若结合足够的训练,识别率可以更高.

Based on the Hidden Markov Model(HMM),a Chinese voice recognition system for nonspecific population is designed,including the recording,training and recognition modules.Specifically,the speech signals with a specific length are recorded.For the recording module,the threshold detection is conducted on short-time energy and zero-crossing rate of voice signals.Meanwhile,the effective voice segments are marked and saved.For the training module,the MFCC(Mel Frequency Cepstrum Coefficient)parameters are calculated using the Baum-Welch algorithm from voice samples to voice templates.For the recognizing module,the similarity probability is analyzed by comparing the voice signals and voice templates via HMM recognition algorithm to detect the maximum output and complete the recognition function.Accordingly,the voice recognition system is complemented using MATLABTM.Therefore,it is found from experimental results that the recognition rate is over 60%,whereas it can be higher through sufficient training.