Processings for speech sound have been carried out in audio band such as spectrographic analysis, autocorrelation, infinite clipping, extremal coding and statistical representation. There is another possibility of processing the frequency shifted speech sigmal, because of the fact that S. S. B. signals contain the same information of the original speech signals. In this paper, we discussed theoretically and experimentally as to how much the essential information can be preserved in this processing. It was found out that this processing can be used for a new information source which has removed the redundancy of phase information with high intelligibility. Based on these results, we have developed speech band compression systems, new extraction systems of the speech parameters such as pitch and formant, and speech recognition system.
The studies concerning the information contained in S. S. B. -clipping of speech has been reported by the authours, in which the information of speech is preserved best by procedure of band-division in each formant region, and the envelope restoration of S. S. B. signal when the phase redundancy can be ignored were also stated. The band-compression system, i. e. Formac, presented in this paper is a system in which information source is an S. S. B. -clipping signal of speech having equivallent spectrum distribution of phase modulation, and the time variation of instantaneous frequency of this signal is extracted in each formant region. Furthermore, this principle has been compared with that of the Vocoder system. From the results of the study, Formac system has been developed which such good results as over 80 % syllable articulation with only one-third of the ordinary speech bandwidth were attained.
We subjectively evaluate the acoustics of rooms and the qualities of acoustical instruments by evaluating musical sounds, speech sounds, and others passing through them. In such a case we had expressed the evaluations by some descriptive adjectives such as "rich", "clear", "resound" and so forth. But there are a variety of expressions to describe the impression of such sounds, and it is uncertain what and how many adjectives we should use. Judging from the number of physical factors influencing those sounds, we expected that it is possible to select a limited number of adjectives sufficiently evaluating those sounds and made two experiments. The first experiment was made from the standpoint of room acoustics. 15 differentlyconditioned sounds made with a reverberation equipment were rated on 21 seven-category scales defined by polar-opposite adjectives. As a result of factor analysis of the intercorrelation between scales by the centroid method, four nearly orthogonal factors were extracted; (1) the aesthetic or lyric factor, (2) the one expressing "volume" and "extent", (3) the one expressing "brightness" and suchlike, (4) the one expressing "softness" and suchlike. The secound experiment was made by varying the lowest resonance of a dynamic loudspeaker by means of a simulator. The ratings were made by the Scheffe's method for pair comparison, and the result proved consistent with the first one.