The frequency modulated tone (FM-tone) plays an important role in the intonation of conversation and the melody of music. Furthermore, it is generally accepted that FM-components in the earlier part of a consonant are essential for speech recognition. There have been, however, very few reports on the sensation produced by FM-tone by mean of psychophysical experiments. In this paper, the sensation of the FM-tone, especialy loudness, was studied by evaluating the masking value. As seen in Fig. 1, a short burst tone (signal) was applied to FM-tone (masker) repeatedly at 5sec-intervals and the critical recognition level of the signal by the normal ear was measured. The signal was superimposed to the masker at various phases so that the masking pattern was obtaind as a function of the time difference between the onset of masker and that of signal. To get the highest time resolution of masking pattern, a 3kHz-signal with 5msec-duration was chosen and the frequency of masker was changed trapezoidally from about 0. 5 to 2kHz with various modulation intervals. The following experiments were carried out : (1) masking patterns produced by ascending and descending FM-tone, (2) monoaural and binaural maskings by FM-tone, (3) effect of modulation interval, (4) masking pattern produced by two FM-tones, (5) simulation of FM-masking, and (6) masking pattern produced by consonant [ra]. From the experimental results, it was concluded that the masking began at about 80msec prior to the onset of the frequency change and increased markedly at the FM-interval and continued 100msec there after (Fig. 3). These phenomena were observed clearly in binaural experiments at both ascending and descending FM-tones (Fig. 4 (a), (b)). In order to confirm this fact, we compared the binaural masking by the FM-tone with that by the alternating AM-tone which was seen in the sonogram of Fig. 5. From this result, it seemed quite clear that the binaural masking depends only on frequency change. This indicated the possibility of the effect of FM-neurons found in the superior olivary on the binaural FM-masking. The masking value depended on also the modulation time. As seen in Fig. 6 the masking value increased markedly in the range of 10 to 100 msec modulation interval for both ascending and descending frequency changes. It should be noted that the modulation time of 10〜100msec is almost the same as that of FM-components in the earlier part of consonant. In the case of consonant, two or three FM-components are present simultaneously or successively. As seen in Figs. 7 and 8, when two FM-tones were applied simultaneously or successively, a masking pattern was obtaind almost as the sum of two independent masking patterns due to respective FM-components. From the results described above, we derived an empirical equation which was applicable in the binaural FM-masking within 0. 5 to 2kHz frequency change. This was a frequency change/df/dt/of the first order linear system and its time constant was about 0. 1sec. Next, in place of FM-tone, we used a voice [ra] which was filtered and gated as seen in a sonogram in Fig. 9 (a). As is evident from Fig. 9 (b), the masking pattern of [ra] increased markedly at the formant transition and separated clearly two formant transitions. The masking pattern seemed to be a valuable method to analyzed the vocal feature extracting process.
