Andreeva N G, Kulikov G A
St. Petersburg State University, 199034, 7/9 University Emb., St. Petersburg, Russia.
Ross Fiziol Zh Im I M Sechenova. 2003 Jun;89(6):715-24.
It is well known that for adult speakers positions of the formants are different for various vowels and they can often be predicted for each phoneme, so first two formant frequencies provide the classic acoustic description of adults vowels' quality. But spectral characteristics of the vowels produced by human voice vary over wide range. In particular this is the case of vocalizations with high fundamental frequencies such as infants' vowel-like sounds and sung vowels. So while the specific structure of a vowel determine its acoustic quality in adults' speaking it does not apply to singing. To test the question of fundamental frequencies dependence of vowels' spectral characteristics, the investigation of vowels [a], [u], [i] sung by two professional singers--tenor and countertenor, was performed. It was shown that with increasing F0 vowel-specific F1/F2 feature tends to disappear. At the same time vowel-specific amplitude relationship between two first spectral maxima tends to occur. So in consistence with our previous data on infants we suggest that the vowel' phonetic identity in high fundamental frequencies might be achieved by information of spectral maxima frequencies (including the first maximum, e. g. fundamental frequency in infants and singers) and their amplitude relations.
众所周知，对于成年说话者而言，不同元音的共振峰位置各不相同，而且每个音素的共振峰位置通常是可以预测的，因此前两个共振峰频率提供了对成年元音音质的经典声学描述。但是，人类发声产生的元音的频谱特征在很宽的范围内变化。特别是在具有高基频的发声情况下，例如婴儿的类元音声音和歌唱中的元音。所以，虽然元音的特定结构在成年人说话时决定其声学质量，但这不适用于唱歌。为了测试元音频谱特征对基频的依赖性问题，对两位专业歌手（男高音和男低音）演唱的元音[a]、[u]、[i]进行了研究。结果表明，随着F0的增加，特定于元音的F1/F2特征趋于消失。同时，前两个频谱最大值之间特定于元音的幅度关系趋于出现。因此，与我们之前关于婴儿的数据一致，我们建议在高基频下，元音的语音识别可能是通过频谱最大值频率（包括第一个最大值，例如婴儿和歌手的基频）及其幅度关系的信息来实现的。
