论文部分内容阅读
提出了一种基于BP神经网络的汉语耳语音转换为正常语音的方法。首先提取正常语音、耳语音的共振峰参数,使用BP神经网络训练出耳语音到正常语音共振峰参数的转换模型;然后根据模型求出与耳语音对应的正常语音共振峰参数,采用共振峰合成的方法将耳语音转换为正常语音。实验结果表明:使用该方法转换的正常语音DRT得分为80%,MOS得分为3.5,在可懂度和音质方面均达到了满意的效果。
A method based on BP neural network to convert Chinese whisper to normal speech is proposed. Firstly, the parameters of formant of normal speech and whispered speech are extracted, and the conversion model of speech from normal to speech parameters is trained using BP neural network. Then, the normal speech resonance parameters corresponding to ear speech are obtained by the model, The method converts the whisper to normal speech. The experimental results show that the normal speech DRT converted by this method has a score of 80% and a MOS score of 3.5, which achieves satisfactory results in both intelligibility and sound quality.