基于听觉模型的汉语耳语音声调检测

Perceiving of Tone in Whispered Chinese Based on Auditory Model

从听觉感知出发,分析了听觉外周模型对于语音激励的主要响应过程,采取听神经平均发放率为声调感知线索,提出了一种汉语耳语音声调的识别方法.其理论基础是听神经发放信息是听觉中枢的唯一信息来源,它是对于语音激励中声强、频谱、共振峰等多种特征的综合反应,因此适合用作耳语音的声调特征.采用BP神经网络对大量汉语元音耳语四声样本进行训练、识别,得到65.1%的平均识别率,达到了改善汉语耳语音声调识别效果的目的.

Based on the analysis of the response of a peripheral auditory model for speech stimulation, the average firing rate of auditory nerves is chosen as the cue for whispered tone. Thus a method for whispered Chinese tone perceiving is proposed. The underlying principle is based on the fact that auditory nerve is the only source of information for central auditory system and it responds to several types of acoustic stimulus such as intensity, formant,etc. Therefore the average firing rate of auditory nerves is a suitable characteristic for the tone of whispered speech. The BP artificial neural network was trained by these proposed parameters to achieve tone recognition. Experiments are performed on a lot of Chinese whispered speech data and the average correct rate reaches 65.1%, which shows that the proposed method is effective for improving the performance of whispered Chinese tone perceiving.