非平稳强噪声环境中的音频信号端点检测系统

Audio signal endpoint detection system in non-stationary strong noise environment

为提高音频信号端点识别能力,设计一种非平稳强噪声环境中的音频信号端点检测系统。构建音频信号端点检测硬件单元,利用预处理单元对音频信号进行预加重、分帧以及加窗处理后,端点检测单元在提取处理音频信号的MFCC倒谱距离特征、频带方差特征的基础上,依据动态阈值估计策略确定恰当阈值;通过双特征参数双门限法来实现对音频信号起止点的确定以及语音帧和非语音帧的分离;利用包络确定延时单元,防止噪声段被错误识别为语音段,避免出现拖尾太长问题。实验结果表明,所设计系统可实现非平稳强噪声环境音频信号端点检测,检测误差满足设定要求。

In order to improve the ability of audio signal endpoint recognition,an audio signal endpoint detection system is designed in non-stationary and strong noise environments.A hardware unit for audio signal endpoint detection is constructed,and a preprocessing unit is used to perform pre emphasis,framing,and windowing processing on the audio signal.On the basis of extracting the MFCC cepstral distance feature and frequency band variance feature of the audio signal,the endpoint detection unit can determine the appropriate threshold based on the dynamic threshold estimation strategy.The dual feature parameter and dual threshold method is used to determine the start and end points of the audio signal and separate the speech and non speech frames.The envelope is used to determine the delay unit to prevent the noise segment from being incorrectly recognized as a speech segment and to avoid the problem of too long trailing.The experimental results show that the designed system can reailze the endpoint detection of audio signals in non-stationary strong noise environments,and the detection error can meet the set requirements.