The research of intelligent speech recognition technology has been going on for a long time. However, due to the characteristics of variability, instantness, continuity and dynamic of the speech signal itself, the identification of the speech still has some difficulties when the machine is put in different environments, especially in the noisy environment. In order to improve the recognition accuracy of the noisy speech signal, a commonly used noise estimation algorithm was studied, which was based on the time-averaged algorithm of posterior signal noise ratio. And an improved algorithm of the smoothing factor was brought up on the basis of the previous algorithm. The voice activity detection algorithm and the above two algorithms were simulated under different input signal-noise ratios. The comparative analysis of the operation results shows that the improved algorithm can improve the output segment SNR by 2.1 dB compared with the voice activity detection algorithm, and it can also improve the output segment SNR by 0.5 dB compared with the original time recursive average algorithm. It is indicated that the improved algorithm can effectively improve the quality and intelligibility of the speech signal at low input SNR.%智能语音识别技术的研究已有较长的时间, 但由于语音信号本身所具有的多变性、瞬时性、连续性和动态性的特征, 使得机器在不同的环境尤其是噪声环境中进行语音信号的识别仍具有一定的困难. 为了提高带噪语音信号识别的准确率, 本文研究了一种常用的噪声估计算法, 即基于后验信噪比的时间递归平均算法. 并在此算法的基础上提出了一种对平滑因子的改进算法, 将语音活性检测算法与这两种算法在不同输入信噪比下进行模拟验证. 通过运算结果的对比分析可以看出, 改进后的算法相比于语音活性检测算法最高可以使输出分段SNR提高2.1 dB, 相比于原时间递归平均算法最高可以使输出分段SNR提高0.5 dB, 表明低输入SNR下改进后的算法可以有效提高语音信号的质量和可懂度.
展开▼
