自适应滤波语音增强算法改进及其DSP实现

DSP realization of modified speech enhancement algorithm based on adaptive filters

为提高强噪声环境下语音信号的信噪比,增强语音通信的质量,以DSP为平台,构建一个基于自适应滤波技术的单通道语音增强系统。该系统以TMS320F2812为核心,结合其多通道缓冲串口(McBSP)与扩展音频接口芯片TLV320AIC23实现了语音信号的高速采集及输出;同时,利用箕舌线函数更新自适应滤波步长因子并引入解相关运算进行语音降噪处理,有效改善了传统算法适应性差,收敛速度慢,稳态误差大等问题。实验结果表明该算法降噪性能好,能明显提高语音清晰度,且系统稳定性强。

In order to improve the signal-to-noise ratio of speech signal under strong noise background and enhance the quality of voice communications, a single channel speech enhancement system based on adaptive filtering techniques is established on the DSP-platform. Taking TMS320F2812 as the core, combining its Multi-channel Buffered Serial Port （McBSP） extended audio interface chip TLV320AIC23, speech signal acquisition and output with high speed is realized. Meanwhile, update the step size of adaptive filtering with tongue-like curve and introduce the de-correlation algorithm for speech de-noising, which improves the adaptability and convergence rate and reduces the steady-state error of traditional algorithm. Test results verify the denoising performance of the algorithm and stability of the system, also it improves the speech intelligibility obviously.