改进的基于多特征的语音端点检测方法

An Improved Speech Endpoint Detection Method Based on Cepstrum Distance

噪声环境是语音识别性能下降的原因之一,端点检测作为其关键技术,其性能优劣在某种程度上决定了识别率的高低。提出一种改进的基于倒谱特征的带噪端点检测方法。在传统基于倒谱距离的算法上综合利用短时过零率和短时能量多特征作为最终判决的门限。实验结果表明,该方法计算效率较高,低信噪比下有较好的检测性能。

One of the causes reducing the capacaty of speech recognition （SR） systems is the noisy enviroment. As an important technology in SR, the endpoint detection accuracy determines the SR rates in certain degree. In this article,an improved method based on MFCC for endpoint detection is proposed. Short time energy characteristics and zero crossing rate are integrated as a new feature and used as the foundation of improved traditional endpoint detection method based on MFCC. Experiments show that high computational efficiency and accurate detection can be obtained under the condition of low SNR.