论文部分内容阅读
为提高实时通信中语音端点检测系统的性能,提出了一种基于能量和鉴别信息的端点检测算法。该算法利用帧信号的能量、子带信号的能量等参数,计算该帧信号与噪声帧基于子带能量分布概率的鉴别信息。算法通过利用鉴别信息,能够在包括语音帧在内的所有帧中更新噪声的能量,从而更准确地跟踪噪声能量的变化。实验结果表明:与基于能量的端点检测算法相比,该方法在信噪比变化比较剧烈的情况下仍然能够较准确地进行端点检测,在0~10 dB范围内变化的坦克噪声环境中,准确率比后者提高约24%。
To improve the performance of voice endpoint detection system in real-time communication, an endpoint detection algorithm based on energy and authentication information is proposed. The algorithm uses the energy of the frame signal and the energy of the subband signal to calculate the identification information of the frame and the noise frame based on the energy distribution probability of the subband. By using the authentication information, the algorithm can update the noise energy in all the frames, including the speech frame, so as to track the change of the noise energy more accurately. The experimental results show that compared with the endpoint detection algorithm based on energy, this method can still detect the endpoint more accurately when the signal-to-noise ratio changes more violently. In the tank noise environment with the range of 0 ~ 10 dB, The rate is about 24% higher than the latter.