基于短时平均幅度和HMM的语音识别系统研究

为克服现行语音识别精度不高的缺点,充分利用资源,改进语音识别效率,研究了基于音频波段特征分析的声音检测与分辨方法。方法以不同人对同一字的发音样本中的音频段信号为主要检测组分,研究不同样本的语音特征区别,使用MNP21声音传感器采集音频信号并进行分析。针对不同人的发音样本体系,提出了使用音频波段检测的思路。基于短时平均幅度优化获得音频信号,进而用隐马尔可夫模型进行识别,设计了语音识别系统。实验结果表明：每人采集10组样本训练,针对五人的不同样本进行多次语音区分,准确率达到100%。

Study on Speech Recognition System Based on Short-time Average Amplitude and HMM

The sound detection and resolution method based on audio-band characteristics was studied to overcome poor recognition accuracy and to make full use of resources to improve the efficiency of speech recognition,i.e.this method detects the phonetic feature of different people whose audio segment signals in the same pronunciation of the word were taken as the main detection component,and through MNP21 sound sensor,having their audio segment signals analyzed;and aiming at the sample system of the different pronunciation,the idea of using audio-band detection was proposed to have optimization of short-term average amplitude based to obtain audio signal for further recognition with hidden Markov model.The experimental results show that this system can achieve a recognition accuracy rate of 100% while carrying out speech recognition of 5 samples of 10 ones collected for training.