Stassen H H
Psychiatric University Hospital, Zurich, Switzerland.
Methods Inf Med. 1991;30(1):44-52.
Motivated by psychiatric interests and as part of our investigations into the basic properties of human speech, we carried out a normative study with 192 healthy subjects--stratified according to sex, age and education--in order to derive reference values of the general population and to learn to distinguish between normal fluctuations and significant changes over time. In the present investigation, our interest focused on the individual sound characteristics of speakers ("timbre") rather than on speech behavior. Accordingly, we determined the optimum parameter setting for a problem-specific, reliable estimation of time-dependent spectra. An interval of one second length was found to be optimum for reproducibly assessing formants and corresponding bandwidths for more than 95% of the cases. Based on these findings, we adapted the concept of "spectral patterns" to speech analysis. It turned out that spectral voice patterns are stable over time and measure the fine graduations of mutual differences between human voices. A highly reliable computerized recognition of persons was possible by means of these quantities, on the basis of 16-32 s time series: 93% of persons could be uniquely recognized after a 14-day interval. Hence, we succeeded in developing specific means for modelling intra-individual changes of voice timbres over time. This is of particular interest for investigations of the speech characteristics of affectively disturbed patients, since the tonal expressiveness of human voices, or the lack thereof, essentially depends on the actual distribution of overtones and the corresponding variabilities.
出于对精神病学的兴趣，并作为我们对人类语音基本特性研究的一部分，我们对192名健康受试者进行了一项规范性研究——根据性别、年龄和教育程度进行分层——以便得出一般人群的参考值，并学会区分正常波动和随时间的显著变化。在本研究中，我们的兴趣集中在说话者的个体声音特征（“音色”）而非言语行为上。因此，我们确定了针对特定问题、可靠估计随时间变化频谱的最佳参数设置。结果发现，对于超过95%的情况，一秒时长的间隔最适合可重复地评估共振峰和相应带宽。基于这些发现，我们将“频谱模式”的概念应用于语音分析。结果表明，语音频谱模式随时间稳定，可衡量人类声音之间相互差异的细微程度。基于16 - 32秒的时间序列，借助这些量可以对人进行高度可靠的计算机识别：14天间隔后，93%的人能够被唯一识别。因此，我们成功开发出了用于模拟个体声音音色随时间变化的特定方法。这对于情感障碍患者语音特征的研究尤为重要，因为人类声音的音调表现力，或缺乏这种表现力，本质上取决于泛音的实际分布和相应的变异性。
