Zero crossing intervals of Japanese monosyllables were measured within accuracy of 25μs, and their time series which contained all informations of clipped speech were represented. Each voice is differenciated before clipping, and the clipped level is set just above noise level. By these operations the articulation of clipped speech will be improved. The time series shows the difference between the clipped speech and the sonagram, and was assumed useful in study of the character of clipped speech, the cause of confusion, synthesis and recognition of voices. Especially, the clipped speech emphasizes the most important parts of voices, and catches up with any rapid variation like turbulent sounds. For example, the time series represent the positions and transition of the hubs or formants of turbulent and unturbulent sounds, and, to some extent, resemble with their sonagrams but they are generally simplified. This article shows the method of measurement, some parts of time series of Japanese vowels and their distributions of zero crossing intervals which have only one or two dominant peaks. And the constructive patterns and their movement of each period of vowels are made clear. In following articles the time series of all Japanese consonants will be represented.
Continued from the preceding article, the time series of clipped Japanese voiceless stop and fricative sounds are represented and compared with sonagrams and other data. As a result, it was recognized that clipped speech, to some extent, holds the important informations of hub of consonants, F_2 locus of following vowels, etc. . The sonagrams of stop and fricative sounds have the characteristic feature of spike fill or fill, but in clipped speech differentiated before clipping, by the character which dominant parts are emphasized, spike fill or fill generally do not appear and show distinctive feature like formant. The sub and F_2 locus of |p| are influenced by following vowels and, to some extent, give actual proof of the results obtained by vocal tract analog by Stevens and House. Mishearing may be due to the transition of following vowels as well as the distribution of turbulent parts and each cause of mishearing in clipped speech can be made clear. Turbulent parts of some stop as well as fricative sounds have time variation in the distribution of their zero crossing intervals.
The first part of this paper describes the significance of speech generation using a digital computer, in relation to the conventional speech analysis by synthesis method, speech recognition, especially that of active or dynamic manner, and on the future speech generating machine. Followed by are the proposals of four methods of constructing a speech generating algorithm workable in computer run; the first has to do with the evaluation of a Fahltung type integral describing human speech generating process, the second, by Fourier inversion of the speech spectrum, the third, by the solution of the differential equation describing human speech generating process, and the fourth, by generating the zero-crossing intervals by pre-defined law. The above algorithms have been investigated in view of time scale of speech generation and precision. In the latter part of the paper, the result of speech generating experiment using the Electrotechnical Laboratory's digital computer, ETL Mark 4-A, is stated, in which, five stationary Japanese vowels have been generated by the approximate evaluation of the Fahltung type integral utilizing simple weighted, sum method. This report is a part of research efforts made under the Laboratory's research project "LOGOS" on linguistic automaton.
Temporary threshold shift was observed in normal persons at high frequencies above 10 kc when the sustained pure tones were applied. The experimental method was as follows: The threshold of hearing was measured at first by routine audiometry at frequencies ranging from 1 kcs to 14 kcs. When the sustained pure tone became inaudible the intensity of the tone was increased successively by a step of 5 dB. This procedure was repeated until the applied tone become audible over 1 minute. At the frequencies above 10 kcs, the threshold shifts were found in 6 cases out of 9. Particularly in some cases the value of the threshold shifts showed 30 dB during 30 second interval. It was found that this phenomenon was due to the auditory adaptation and not to fatigue. The reason was that when the tone was interrupted, the threshold shift instantly disappeared and did not persisted as in the cases of fatigue. Concerning the origin of this phenomenon, two reasons were considered. First, at high frequencies, the sound pressure of minimal audible threshold was higher than that at middle frequencies. Secondly, the vibrating portion of the basilar membrane in the cochlea became narrower in proportion to frequencies.
A variational method for treating the diffraction of sound is applied to the case of an infinitely thin rigid rectangular plate, for which no exact solution is known at present. Utilizing the integral equation, the diffraction coefficient is expressed in a form which is stationary with respect to small variations (relative to the correct values) of the difference in velocity potential on opposite plate faces. And numerical calculations of the diffraction coefficient and the radiation pressure on the plate are made for a/b=1, 0. 5, and 0. 25, where 2a and 2b are the lengths of the sides respectively.
Measurements of the vibration acceleration at one point of the plate and that of the sound pressure near the plate surface were made in succession with the preceding report of the same title. The stiffness of rubber hammer were measured and the impulsive force figures were observed by using piezo-electric materials and a synchroscope. Attention was given to the coincidence frequency which makes the wave length of the sound in air equal to that of the flexural vibration in the plate, as well as to the force figure, and following results were obtained: 1) The correspondence between vibration acceleration and sound pressure is fairly complex depending upon the properties of materials and its thickness. As to the radiation of sound from a plate, the coincidence effect could be observed, but not so remarkable as expected from the theory. 2) By reducing stiffness of the hammer, the components of high frequency of sound pressure and vibration acceleration tends to disappear, but on the contrary, not so much difference can be observed in a low frequency region. The force figure decides the frequency response and it varies according to hammer stiffness, weight and velocity.
Sonic agglomeration of aerosols (tabacco smoke and NH_4 Cl smoke) was investigated in progressive and standing sound wave fields of frequency ranging from 0. 5 to 10 kcs/sec in a glass tube. Falling time was determined by measuring the intensity of the transmitted light in the natural state and in sound fields. At the same time, microscopic examination of the particle size was made on samples collected on glass plates. It was revealed that the particle size of the collected samples increased as the falling time decreased or vise versa, both in the natural state or in sound fields. Sound waves shortened the falling time appreciably, and the effect became greater, as the sound pressure was increased. Optimum frequency was observed for each species of the aerosols according to their particle sizes. This optimum frequency was almost independent of the sound pressure and the concentration of the aerosols. The effect was nearly the same in standing waves as in the progressives, provided the sound energy density was common in both cases.
The author has developed a thin film thermo-couple made by evaporating process in combination with a D. C. galvanometer to function as a detector or a device for measuring intensity of sound field. The author calculated the temperature rise of the thermo-couple assuming that the mechanism is a kind of hysteresis loss caused by the delay of alternating heat conduction between air medium and rigid surface in analogy to Dr. J. Saneyoshi's theory on the heat conduction damping of pulsating gas bubbles which resonate to ultrasonic wave in liquid. It is proved that the value of temperature rise measured by the thermo-couple is coincident with the value theoretically calculated. The temperature rise is proportional to the square of the sound pressure and to the square root of the sound frequency according to this theory. The fact that this detector has the same characters is also ascertained by this experiment.
The characters of the long Langevin type transducers are already analyzed, but the elastic constants of the adhesive layer, to be used in this theory, are not clarified sufficiently. The writer made a study of this problem. It is assumed in this study, that the lateral displacement of the adhesive layer is limited by the displacements of the metallic part and the electro-strictive layer by continuity requirements. As a result, it was clarified that the wave type to be applied for the adhesive layer is a type close to the bulk wave. It is confirmed that the theoretical values introduced from this treatment are in good agreement with the experimental values. Consequently the effective elastic constants for the adhesive layer in the Langevin type transducers were clarified.