This paper presents a method of adaptive digital filtering to remove the noise components from the noise-added speech in order to improve its articulation. Fig. 5 shows the flow chart of this filtering process which is adaptive and nonlinear. By FFT, the filtering is done only in the frequency domain. The system function of the filter is determined from a criterion to reproduce the amplitude-spectrum of speech (Eq. (3)). The left path of the flow chart is the path for the input which is to be processed. The right path is the process to determine the system function of the filter, which is determined by the amplitude-spectrum of the input itself. The processing to extract speech components is illustrated on the right side of Fig. 5. A threshold is used to remove the noise component whose frequency-range is different from that of speech. Fig. 7. illustrates how the real data is transformed at each stage of the processing. From the syllable articulation test, it became clear that the glide part of speech must be processed specially to improve the articulation scores. Consideration of this point improved the processing. Each input frame is segmented depending on the unvoiced sound, the glide or the stationary vowel. The system function of the filter is determined differently according to this segmentation. Fig. 8 is the flow chart of this improved processing. The syllable articulation scores were improved by about 8%, both in the case of white and pink noise, by this improved processing (Table 2).
The longitudinal displacements of a point on a piano string changing with time immediately after hammering are calculated, together with the transverse displacements, using the finite element method. This method assumes that the string is elastic and completely flexible with small mass points and a large mass point of the sound board on it. From equations (1)〜(6) and Fig. 1, the values of transverse and longitudinal displacements changing with time at any point of the string immediately after hammering may be calculated (Figs. 2, 3, 4). It is found from the analysis that the longitudinal vibration generated in the string causes the sound board to produce an inharmonic tone of high frequency (Fig. 5).
"Vibrato" is a certain periodic variation of sound properties. It is classified into three kinds : frequency vibrato, amplitude vibrato, and spectrum vibrato. The relations between hearing and various factors of frequency vibrato were investigated and the fundamental frequency (f_0) of each sound stimulus used was 440 Hz in this investigation. Initially the principal pitch and pleasantness of the FM sounds were examined as a function of the rate of vibrato (p_0) and extent of vibrato (2×Δf) in the case of pure tone. From the results of the research on principal pitch, the relation m=F(p_0/Δf) was obtained (Fig. 4). This was the same as the result obtained by K. Hirose (Table 2). However, when Δfwas small, a different tendency was observed (Fig. 5). From result of the study on the pleasantness of pure tone with frequency vibrato, the average pleasantness was maximum at p_0=7 times/sec and Δf=0〜2 Hz. The smaller or larger than 7 times/sec p_0 or the wider Δf was, the lower the average pleasantness became (Fig. 8, Fig. 9) . In comparison with Seashore's research on frequency vibrato of the tones of music performances, our results were the same with respect to p_0 but different with respect to Δf. Seashore demonstrated that the maximum preference of frequency vibrato was observed when Δf was about a quarter-tone in singing voices, and one-eighth in musical instruments like the violin. In the following investigation, p_0 was fixed at 7 times/sec. The next study concerned the relation between Δf and the pleasantness of complex FM sound with 5 harmonics whose spectrum envelope shapes were flat. The average pleasantness was maximum when Δf was 0〜1 Hz, and the wider Δf was, the lower the average pleasantness became as in the case of pure tone (Fig. 9). When we changed the shapes of the spectrum envelope of the complex FM sounds with 5 harmonics, we found a tendency for the average pleasantness to become higher as the shape became similar to pure tone (Fig. 10). Finally, from the results of the study on phase inversions among the AM signal waves of the 5 harmonics of a complex sound, the differences of this phase inversion could be detected and the average pleasantness was maximum in the case of the sound stimulus whose phases of the AM signal waves of its harmonics were made inverse between odd and even harmonics (Fig. 11). A paired comparison test was used for the measurement of pleasantness. These results will not be directly applied to the tones of music performances, but they will be useful in the case of designing sound signals, etc. .