室内声环境中的语音端点检测自适应算法

室内场所是语音识别技术的一种典型应用环境,传统的端点检测研究多考虑噪声的影响,忽略室内混响的影响,研究证明室内混响对端点检测和识别效果能造成显著的负面影响.通过研究短时能量和短时自相关序列( RAS),提出了一种自适应的端点检测方法.可以通过估计噪声段短时能量来适应平稳噪声干扰环境,并能修正含混响语音的检测终点.端点检测和语音识别实验结果表明,本方法在平稳噪声和室内混响声环境下具有良好的性能.

Adaptive Algorithm for Speech Endpoint Detection in Sound Environment in Rooms

Enclosed space is a typical application environment for the technique of speech recognition. Traditional studies in the speech endpoint detection usually consider the white noise environment, but rarely the room reverberation case. In the sound fields in rooms the correct rate of isolated word recognition degrades dramatically. One of the most important reason is the ineffective speech endpoint detection algorithm. A new adaptive detection method based on the short - term energy and relative autocorrelation sequences （RAS） is presented. This method eliminates the noise energy and can work in an effective and adoptive way in white noise circumstance, and can correct the end point of the reverberant speech. The experimental re- suits are based on endpoint detection and recognization system demonstrate that this method performs well when the speech is affected by the white noise and the room reverberation.