非平稳强噪声环境中的音频信号端点检测系统OA北大核心CSTPCD

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.