FROM CURRENT LITERATURE

A NEW ELECTRICAL DEVICE FOR THE REPRODUCTION OF VOWELS*

In a report read at the Prussian Academy of Sciences, the well-known German electrical engineer K. W. Wagner reports on the results of extensive work on the analysis and synthesis of vowels, carried out in his laboratory.

Wagner points out that the work of a number of investigators who in recent years have experimented with very refined apparatus has confirmed the basic ideas of Helmholtz’s theory; namely, it has been established with complete certainty that:

1) every vowel sound corresponds to a certain periodic oscillation;

2) the specific character of a vowel depends on the presence in its spectrum of characteristic additional frequencies (formants), the position of which is almost independent of the pitch at which a given vowel is pronounced; the pitch affects only the relations between the amplitudes;

3) the subjective sensation of a sound does not depend on the phase shifts between the separate frequencies entering into the spectrum of the vowel (G. Ohm’s law).

Work in recent years has shown that a vowel is characterized not only by the two principal formants, whose existence has long been known, but by three or four. The additional formants are expressed more weakly than the main ones, but their establishment is nevertheless of full doubtlessness.

Attempts to synthesize vowel sounds, going back to Helmholtz himself, have on the whole yielded positive results; however, complete agreement between the timbres of a natural and an artificial sound has been achieved only in a few particular cases.

The best results in this direction were obtained by Stumpf, who used for the synthesis of vowels 28 whistles, the sounds of which were carefully freed of overtones and then mixed.

On the other hand, a number of attempts are known at the direct creation of vowel sounds with the aid of mechanical models reproducing the human vocal apparatus.

The functioning of the vocal apparatus, as Helmholtz already indicated, reduces to the following: the larynx, serving as the exciter of sound vibrations, is connected through the vocal cords with the palate, nasal, and oral cavities, which play the role of resonators. By varying the shape and size of the mouth and by positioning the tongue in a definite way, one can change both the proper periods of the cavities and the relations between them.

The role of the vocal cords consists in the fact that, depending on their tension, they pass more or less frequent impulses of the air stream from the larynx.

* Wagner, Abhandl. d. Preussischen. Akad. d. Wissenschaften, Phys.-mathem. Klasse, 1936 (Sonderabdruck).

Thus, in the vocal apparatus we encounter impact excitation of resonating systems. The frequency of the impact impulses determines the fundamental pitch of the sound; the natural periods of the resonators and the relations among them, as well as the degree of damping of each of them, determine the formation of one or another formant.

In this connection, the damping of the resonators plays a very important role, clearly illustrated by Fig. 1.

Fig. 1.

Short-duration impact impulses, whose spectrum is shown at the upper right, act on a resonating system possessing two natural frequencies, \(f_1\) and \(f_2\). In the left-hand part are shown the oscillations arising in the system (thin lines) and the resulting curve (thick line), and on the right—the spectrum of the resulting oscillation under strong and weak damping. The emergence of two formant regions is revealed quite distinctly (especially with weak damping).

At present, while we have fairly detailed information about the dimensions of the resonating cavities and their natural frequencies, there is almost no information about their damping.

Fig. 2.

Mechanical models of the vocal apparatus have been constructed repeatedly and in individual cases have yielded satisfactory results. However, their substantial shortcoming was the circumstance that for each vowel it was necessary to create a separate model.

In 1922 an electric model was proposed, the circuit of which is shown in Fig. 2. A generator producing oscillations with a large number of overtones feeds two oscillatory circuits, whose frequencies and dampings are easily varied by changing the inductances, capacitances, and resis-

...phenomena. The resulting oscillations are picked up by a telephone and converted by it into sounds. In this way it has proved possible to reproduce a number of vowel sounds quite satisfactorily.

The apparatus for reproducing vowels, proposed and constructed in Wagner’s laboratory, is built on the same principle, but is made more perfectly. Its schematic diagram is shown in Fig. 3. The direct-current source plays the role of the lungs (the supply of energy). The impulse oscillator \(SO\) is equivalent to the larynx, the circuit \(G\), which creates the fundamental tone, to the vocal cords, the circuits \(F_1—F_4\) to the resonating cavities of the throat, mouth, and nose, and the loudspeaker to the opening of the mouth.

Fig. 3.

The impulse oscillator (a circuit with a dynatron, making it possible to obtain intermittent oscillations of short duration with relatively large pauses between them) produced pulses rich in intense high overtones (up to the 30th order).

Fig. 4.

The circuit \(G\) is an oscillatory circuit tuned to the desired fundamental frequency. The formant circuits \(F_1—F_4\) are oscillatory circuits tuned ...

to the desired formant frequencies; the damping of the circuits could be varied over wide limits. The filters not shown in the diagram made it possible to eliminate the mutual influence of all the circuits upon one another and upon the shock oscillator, so that each formant could be adjusted independently of the others. Circuits \(G\) and \(F\) were closed to potentiometers, from which a voltage was taken that was fed to the loudspeaker reproducing the corresponding sound. This device made it possible to reproduce vowels with all the individual features inherent in the voice of one or another person.

For this purpose an oscillogram of a given vowel was recorded and a spectral analysis of the oscillogram was carried out. According to the data of the analysis, the corresponding frequencies and dampings of the formant circuits were set.

Analysis of a series of curves of natural sound showed that the spectral characteristics of successive periods are not entirely identical; if one and the same person pronounces a vowel several times in succession (with intervals), then the spectral characteristics differ still more significantly, although a certain commonality among them is preserved. Therefore Wagner considers it expedient to analyze a series of periods and to find an average spectral characteristic, according to which the apparatus reproducing the sound is then adjusted.

The results of the reproduction proved to be exceptionally successful. By subjective sensation it is impossible to distinguish the natural sound from the artificial one. The physical characteristic of the quality of the reproduced sound is illustrated in Fig. 4, in which the upper oscillogram and its spectrum correspond to the natural sound, and the lower one to the reproduced sound.

Fig. 5.

The spectral distributions almost coincide; the discrepancy in the shape of the curves on the oscillograms is of no significance, since it is due to the difference in phase shifts, which do not affect the subjective sensation of the sound. These curves characterize the German “a,” sung at a fundamental tone of 117 Hz.

Fig. 5 makes it possible to compare the natural and reproduced sound sung at a fundamental tone of 156 Hz (the German “u”).

The value of the apparatus proposed by Wagner lies not only in the possibility of easily reproducing the individual timbre of the voice, but also in the possibility of studying the influence on the timbre of the sound of individual formants, since the change in amplitude, damping, and frequency of each of the four

of the formant circuits can be carried out quite independently of the other circuits of the apparatus; individual formant circuits can even be completely disconnected from the circuit.

The importance of such investigations for the study of the human vocal apparatus and for the development of methods of combating speech defects can hardly be overestimated.

N. Malov, Moscow