双传声器近讲语音增强算法的DSP实现

Realization of Close-talk Speech Enhancement Algorithm Based on DSP

针对近讲模式,提出了一种可以DSP实现的双传声器语音增强算法,算法在安森美公司的BelaSigna 300 DSP平台上进行了实时实现。介绍了该算法的基本原理、BelaSigna 300 DSP系统的硬件结构、其软件编写方式及移植到DSP系统上后的算法处理流程和移植过程中的关键点分析。在白噪声、音乐噪声与广播噪声3种噪声环境下进行了实验,并从语谱图、MOS、SNR三个角度将该方法与谱减法、维纳滤波、MMSE进行了比较。实验结果表明,该算法对于3种类型的带噪语音,信噪比均可以最多提高20dB,提高后的信噪比在30dB以上,并能较好地保持目标语音的语音质量。

Aiming at close-talk applications, a two-mie speech enhancement algorithm is proposed and realized based on BelaSigna 300 DSP system of On Semiconductor. Principle of the algorithm, hardware structure and software platform are introduced. Then real-time processing flow is illustrated and key points of transplanting are listed. In the end, experiments are conducted under three kinds of background noise situations and white noise, musical noise and broadcast noise are ehoosed. Results are compared with that of Spectral Subtraction, Wiener Filtering and Minimum Mean Square Error and observed from perspectives of spectrogram, Mean Option Score and SNR. Experiment results show that the algorithm surpasses other three methods and can increase SNR of three kinds of noisy signals by 20 dB at most and to more than 30 dB and meanwhile keep good qulity.