实时语音采集系统中语音端点检测和增强方法

Research on Endpoint Detection and Speech Enhancement in Real-time Voice Recording System

针对实时语音采集系统应用到工业行业中所面对的增加有效记录时间和提高噪声环境下的录音回放效果问题,研究了端点检测方法和语音增强方法。端点检测方法基于有限状态机模型,引入通信中AMR-WB标准的激活检测算法作为语音帧或噪声帧判据,端点检测可达到良好的实时性和准确性,并对于低幅度信号增加了过零率判据可进一步降低检测算法对语音的漏报率。语音增强方法以参数可配置的低频带阻滤波结合维纳滤波来增强语音采集系统对多种环境噪声的消除能力。实际使用表明该端点检测方法和语音增强方法对于提高工业领域应用中实时语音采集系统的记录长度和回放效果具有很强的实用性。

The real - time voice recording systems using in industrial applications are facing to increase effective recording time and acquire enhanced playback quality in noisy environment, and voice endpoint detection and speech enhancement al- gorithm for these purposes are focused on. The voice endpoint detection is based on finite state machine architecture and the AMR - WB standard in mobile communication is introduced as the main un - voice/voice detector, zero - crossing rate is al- so used for detecting very low amplitude signal. The speech enhancement combines configurable low - cut filtering and Wie- ner filtering for adapting different kind of environment noises. The applying in traffic dispatch monitor production shows that these voice endpoint detection and speech enhancement algorithm are very practical in extending recording time and reducing playback noise.