【摘要】 提出一种噪声下的多数据流子带语音识别方法。传统的子带特征方法虽然能提高噪声下的语音识别性能,但通常会使无噪声情况下的识别性能下降。新方法提取感知线性预测(PLP)特征和子带特征,分别进行识别,然后在识别概率层将两者相结合。通过E-Set在NoiseX92下的白噪声的识别实验表明,新方法不仅具有更好的抗噪性能,而且同时能提高无噪声情况下的识别性能。更多还原

【Abstract】 In this paper,we present a multi-stream sub-band method for noisy speech recognition.The conventional sub-band methods can improve the recognition accuracy of noisy speech,but degrade that of clean speech.In the new method,PLP feature and sub-band feature are extracted,then combined at probabilistic layer.Evaluated by a task on E-SET database under white noise from NoiseX92 noise database,the new method can increase the recognition rate for both noisy speech and clean speech.更多还原