一种基于噪声模型的语音/噪声分类方法

A Noise Model Based Method for Speech/Noise Discrimination

提出了一种可用于嵌入式ASR系统的语音/噪声分类的新方法,该方法利用一个噪声模型,对每帧信号的评价值进行语音/噪声分类.实验表明,该方法可以有效地区分语音和噪声,并表现出在各种噪声环境和不同信噪比条件下的鲁棒性.该算法已经集成进一个ASR系统,并在Com-paqiPAQ上进行了测试,其计算代价不到整个系统代价的10%.

A robust method suitable for embedded automatic speech recognition(ASR) system was proposed to distinguish speech signals from noise signals. The method is based on a noise model, according to which a score is used to distinguish speech from noise. The experiments show that the method is effective in speech/noise discrimination and robust in various environments at any SNR. The algorithm has been employed as the front-end of an ASR system on Compaq iPAQ, and its cost is less than 10% of the ASR system.