环境噪声是语音识别和说话人识别性能下降的原因之一,端点检测作为其关键技术之一,性能优劣在某种程度上决定了识别率的高低。文章提出一种改进的基于倒谱特征的带噪语音端点检测方法。在传统基于倒谱距离的算法基础上,该算法进一步综合利用短时过零率和短时能量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。 Environmental noise is one of the reasons for the degradation of speech recognition and speaker recognition performance. Endpoint detection, as one of its key technologies, determines the recognition rate to a certain extent. This paper presents an improved cepstral feature-based endpoint detection method with noisy speech. Based on the traditional algorithm based on cepstrum distance, the algorithm further uses the short-term zero-crossing rate and short-time energy as the final decision threshold. Experimental results show that this method has higher computational efficiency and better detection performance under low SNR.