残響による音声の品質劣化についての一考察

抄録

One of the typical interference factors which should be taken into account on using loud speaker telephone is the deterioration of speech quality caused by reverberation. To investigate such effects of reverberation, the speech qualities of artifical and real room reverberation were estimated from the viewpoint of preference. In order to simulate the simple phenomenon of reverberation in this experiment, the colorless artifical reverberation equipment proposed by Schroedere was composed with a digital and an analogue computers. Because the reverberation equipment has the simple feedback system comprising three amplifiers and one delay element (Fig. 1), it is easy to change some physical parameters of reverberation. For subjective measurement, preference tests were carried out by the pair comparison of two signals. The one is speech with reverberant and the other is the original speech added with white noise. From these results, the signal to noise ratio at 50% preference point, defined as equi-preference signal to noise ratio, was determind. In order to remove the effect caused by the different delay time of artifical reverberation, the normalization was carried out. The equi-preference signal to noise ratio of speech with artifical reverberation and reverberation speech in real room is shown in Fig. 4. According to the results of these experiments, the quality of reverberant speech can be expressed by a signal straight line in the region of reverberation time T_r shorter than about 250 or 300 msecs. Therefore the relation equi-preference signal to noise ratio with T_r is shown by the equation (2). Under the condition that the impulse response representing reverberation decays exponentially with time, it can be rewritten as the equation (7). The equation (7) is similar to that giving the Definition or Clearness by Mayer, that is such value that the energy in first 50 msec of the impulse response is divided by the total energy. Therefore the preferable signal seems to be the total energy of direct sound and the following sound. On the other hand, it is considered that the energy of sound coming about 40msec after direct sound acts as interfering noise in view of preference for telephone transmission system. In order to cerify this supposition, other preference tests were carried out by the pair comparison of speech followed by echo and noise-added speech. The result shown in Fig. 8 suggest that the echoes coming after about 40 to 50 msec cause the same deterioration of speech quality except the case of single echo. From these consideration, it may be useful, in the viewpoint of preference, to suppress the component of sound energy coming later than about 50 msec for telephone transmission system.