有效提取耳语音共振峰的改进方法

Improved method for effectively extracting whisper speech formant

耳语音是噪声源激励,与正常音相比,其共振峰位置发生了偏移,带宽增宽。故采用传统的线性预测法提取耳语音共振峰时存在虚假峰问题。通过分析功率谱,提出了一种改进算法。根据极点功率不变的原则,利用极点交互因子修正共振峰的带宽,从而准确地提取出耳语音的共振峰。对汉语普通话单元音音素仿真实验的结果证明了该算法的有效性。

Whisper is stirred by noise.Comparing with normal speech,the formant of whisper is shifted and the bandwidth is broadened,and that will bring up the problem of spurious peaks when using the tranditional conventional liner prediction coding for formant extraction.By analyzing power spectrum,an improved approach has been proposed.Based on the rule that the pole power is not change,the algorithm modifies the whisper formant bandwidths using pole interaction factor,and extracts formants exactly.Experimental results with mono-vowel phones in Mandarin speech prove the ability of this algorithm.