针对无线电语音传输过程中低信噪比(Signal-to-Noise Ratio,SNR)的情况,改进优化谱减法,通过对多窗谱谱减算法中参数的自适应选择,获得对噪声频谱的合理估计,提高信号的降噪效果,并使用谱熵为特征的端点检测算法对人声进行提取.仿真对比实验结果显示,在强噪声环境下,所提算法在语音增强方面表现更出色,识别准确率更高.
This paper focuses on the low Signal-to-Noise Ratio(SNR)scenario in the process of radio speech transmission.It enhances and optimizes the spectral subtraction method by adaptively selecting parameters in the multi-window spectral subtraction algorithm.This adaptive selection provides a reasonable estimation of the noise spectrum,thereby improving the denoising effect of the signal.Additionally,a feature-based endpoint detection algorithm using spectral entropy is employed for voice extraction.Simulation and comparative experimental results indicate that,especially in noisy environments,the proposed algorithm outperforms in terms of speech enhancement,achieving a higher recognition accuracy.