Speech Recognition Using Wearable and Environmental Microphones under Real Environments

我们提出了一种新的噪声抑制使用两种麦克风与自动时间同步。其中一个麦克风用于估计噪声信号，并且从另一个麦克风的目标输入信号中减去所估计的噪声分量。利用语音特征距离进行自动时间同步，并通过估计转换矩阵补偿噪声频谱的空间特性。实验结果表明，该方法对非平稳噪声信号的检测更为有效。

We propose a new noise suppression using two kinds of microphones with auto time-synchronization. The one of microphones is used to estimate noise signals, and the estimated noise components are subtracted from the target input signal of the other microphone. Auto time-synchronization is done by using speech features' distance, and the space characteristics of noise spectrum is compensated by an estimated conversion matrix. Experimental results show that this method is more effective for non-stationary noise signals.