论文部分内容阅读
针对在低信噪比条件下难以实现语音端点检测,提出了幅度--带宽联合分析的解决方法,自适应地调整各帧噪声功率谱,有针对性地进行谱减;通过削减随机噪声谱峰,达到了抑制噪声的目的。然后采用语音功率谱点与噪声功率谱点之比序列的方差信息量对语音信号进行双门限检测。新算法经过仿真实验,能够有效地区分语音和噪声,可以显著地提高语音识别系统的性能,在不同的低噪声环境条件下具有鲁棒性。该算法计算代价小,实时性好,简单易实现。
Aiming at the difficulty of voice endpoint detection under low signal-to-noise ratio (SNR) conditions, a solution of amplitude-bandwidth joint analysis is proposed. The noise power spectrum of each frame is adaptively adjusted and the spectral subtraction is performed. By reducing the peak of random noise , To achieve the purpose of suppressing noise. Then the variance information of the speech power spectrum point and the noise power spectral point is used to detect the double threshold of the speech signal. The new algorithm through simulation experiments, can effectively distinguish between voice and noise, can significantly improve the performance of speech recognition system, in different low-noise environments with robustness. The algorithm is less costly, real-time, easy to implement.