论文部分内容阅读
噪声环境下的语音端点检测在稳健语音识别中占有十分重要的地位。自适应子带谱熵法是一种新的端点检测方法,它的思想是将一帧语音分成若干个子带,再用谱熵法进行运算,子带的个数可以自适应选择。该方法具有一定的稳健性,但随着信噪比的降低,语音端点检测的准确性也随之下降。提出了一种结合加权功率谱减的子带自适应谱熵法,并给出了该方法的实现步骤。该方法采用边降噪边用稳健性好的特征参数做语音端点检测,从两个方面来提高端点检测的准确性。实验结果表明该方法具有良好的性能,在不同信噪比的不同加性噪声下系统识别率都有提高。
Voice endpoint detection in noisy environments plays a very important role in robust speech recognition. Adaptive subband spectral entropy method is a new endpoint detection method. Its idea is to divide a speech into a number of subbands and then perform spectral entropy method. The number of subbands can be adaptively selected. This method has a certain degree of robustness, but as the SNR decreases, the accuracy of voice endpoint detection also decreases. A subband adaptive spectral entropy method combined with weighted power spectrum subtraction is proposed and the implementation steps of this method are given. This method uses edge-noise-reduction edge to do voice endpoint detection with good robustness feature parameters, and improves the accuracy of endpoint detection from two aspects. Experimental results show that the proposed method has good performance. The system recognition rate is improved under different additive noise with different signal-to-noise ratios.