低信噪比下语音多路信号端点切分仿真

Speech Multiplexed Signal Endpoint Segmentation Simulation With Low SNR

针对当前语音信号端点切分相关方法存在召回率与准确率低的问题,提出基于短时相对能频积的低信噪比下语音多路信号端点切分方法。依据听觉掩蔽原理,利用语音多路信号端点检测方式在静音阶段得到噪声幅度,采用基本谱减法提升信噪比。计算听觉掩蔽阈值调节谱减系数,并得到语音信号谱评估值,对信号谱评估值进行傅里叶反变换获得增强之后的语音信号。基于增强后的语音信号,引入短时相对能频积,定义并提取语音信号特征参数。根据信号特征参数中的能量、过零率与能频积设置端点检测与切分合理阈值,利用所得阈值分别判断语音信号初始端点和结尾端点,以此实现语音多路信号端点检测切分。实验结果表明,所提方法具备很高的召回率和准确率,综合性能优于当前相关研究成果,鲁棒性强。

At present, the current method leads to low recall rate and low accuracy rate. Therefore, a method to segment the speech multi-channel signal endpoint with low signal-to-noise rate (SNR) based on short-time relative energy-frequency-value is proposed. According to the principle of auditory masking, the multipath signal endpoint detection method was used to obtain the noise amplitude at silent stage, and the basic spectral subtraction algorithm method was used to improve the signal to noise ratio. The auditory masking threshold was calculated to adjust the spectral subtraction coefficient, and then the spectrum evaluation value of speech signal was obtained. Moreover, the inverse Fourier transform was performed on the signal spectrum evaluation value, so that the enhanced speech signal was obtained. Based on the enhanced speech signal, a short-term relative energy-frequency-value was introduced to define and extract the speech signal characteristic parameters. According to the energy, zero-crossing rate and energy -frequency-value in the signal characteristic parameter, the reasonable thresholds of endpoint detection and segmentation were set. Finally, the obtained threshold value was used to judge the initial endpoint and ending point of speech signal respectively. Thus, the endpoint detection segmentation for speech multi-channel signal was achieved. Simulation results show that the proposed method has high recall rate and accuracy rate. Meanwhile, the comprehensive performance is better than that of current research results and the robustness is strong.