In order to analyze the articulation mechanism of a talking bird "Mynah, " mimic voices obtained in He- O_2 atmosphere are compared with those in the normal air based on the frequency analysis by linear prediction. A mixture of 80%-He and 20%-0_2 at one atm is used in the experiment of helium voices, and 36 mimic voices consisted of words are recorded by a taperecorder in both cases. Comparing vowel parts of normal voices with those of helium voices, it is found that the spectrum pattern of mimic voices has two types, I and II, about the structure of spectral envelope peak components. Type I shows the frequency shift in helium atmosphere and type II shows no frequency shift. Spectral envelope peak components of type I appear around 0. 8-1. 2kHz, 1. 8-2. 4kHz and 3. 8-4. 4kHz, and their frequencies do not change with vowels. On the other hand, the component of type II appears in the range from 1. 5 to 3kHz and its frequency changes with vowels. It is ascertained that (1) only the spectral envelope peak components of type I show almost the same frequency shift as those estimated by the formula describing theoretical transformation of peak frequencies, (2) spectral envelope peak frequencies of type I coincide with resonance frequencies obtained by twin-tube model of which parameters correspond to the vocal tract of Mynah, (3) the number of observed spectral envelope peaks are beyond that of the resonances calculated by twin-tube model, and (4) the observed spectral envelope coincides with one which is obtained by adding a component of type II to the spectral envelope estimated from the source-system model using spectral envelope peaks of type I as poles. From above results and analyses, it is confirmed that type I has multi peak components which are formed by resonances of the vocal tract, and type II has a single peak component which may not be formed by the resonance. Furthermore, it is assumed that Mynah's syrinx may produce two components as sound sources. The first one has a wide range spectrum composed of pitch and its harmonics, which is similar to the sound source by human's vocal cord. The second one has a higher frequency component corresponding to type II. This hypothesis may be also supported by the anatomical structure of the syrinx. From above considerations, it is concluded that Mynah may produce the vowel part of mimic voices by controlling the frequency of the second sound source in addition to use almost fixed resonance components obtained by its vocal tract.

抄録全体を表示

In order to make it clear what kind of physical characteristics mimic voices of a talking bird mynah has, 111 mimic of a well-trained mynah are analyzed in (1) the intensity pattern of sound wave form, (2) the fundamental frequency contour, (3) the long term spectrum, (4) the spectral envelope peaks of vowels, (5) the time spectral pattern of consonants. And they are compared with those of tutor's voices. (1) Though the intensity patterns of mimic voices are a little bit different from the tutor's, the duration of the pause in the speech sound and the duration of the burst in some stop consonants are very similar to those of tutor's voices. (2) The fundamental frequency contour shows generally good coincidence between them. The fluctuations of fundamental frequency are also found in mimic voices, however, they are comparatively larger than those of tutor's voices. (3) The long term spectrum of mimic vices has a peak around 900 Hz and the slope for the lower frequency part is 2 dB/oct. and -10dB/oct. for higher frequency part. So, the spectral components below 500 Hz are smaller than those of 1 to 3 kHz by 3 to 10dB. (4) As to the first spectral envelope peaks (P_1) appear around 1kHz in every vowel, it seems very hard to categorize five vowels on the P_1-P_2 plane. (5) As to the spectral changes, the transition of the spectral envelope peaks in the range of 1 to 3kHz are found similar to the tutor's though that of P_1 is not so clear. These physical characteristics of mimic voices of mynah bird are supposed to depend not only on its vocalization but also its auditory function, so that some suggestions will be obtained for the studies of speech synthesis and automatic speech recognition. However, some questions still remain as to how mimic voices with different spectral pattern are processed in the human auditory system.

抄録全体を表示