A speaker identity perception for formant-suppressed vowels

在本研究中，我们利用共振峰抑制元音刺激来研究说话人身份认同的知觉机制。首先，对300名成年男性和女性说话者发出的日语元音进行声学分析。结果表明，全谱特征比已知的形成峰频率具有更多的信息。在此基础上，合成了共振峰抑制的元音刺激。利用这些刺激进行知觉实验。在实验中，三个元音（ABX）被依次呈现给听者。听者被要求回答同一个说话者对x发出的元音A或B。结果表明，基本频率和形成峰频率都是说话者身份感知的有效线索。然而，结果也显示了关键的听众依赖性。因此，很难概括说话人认同的知觉机制。

In the present study, perceptual mechanism for speaker identity was investigated using formant-suppressed vowel stimuli. At first, Japanese vowels were acoustically analyzed which uttered by 300 adult male and female speakers. The results indicated that whole-spectral characteristics had more information than well-known formant frequencies. Based on the analysis, formant-suppressed vowel stimuli were synthesized. Using the stimuli, perceptual experiments were carried out. In the experiments, three vowels (ABX) were successively presented to listeners. The listeners were required to answer vowel A or B, which was uttered by the same speaker for vowel X. The results indicated that both fundamental and formant frequencies to be effective cues for speaker identity perception. However, the results also showed crucial listener dependencies. Thus, it was difficult to generalize perceptual mechanism for speaker identity.