定レベル交叉波の特質とその応用 (<音声特集号>)

抄録

The infinitely clipped speech investigated by Licklider and others shows that the information of speech is sufficiently contained in that of zero-crossing points of speech wave. From this point, a new voice communication system in which each position of zero-crossing is transmitted by pulse is thought out. However, the infinitely clipped speech itself can not be used for this purpose, but it is necessary to depress the noise in intervals of speech pause and to definitely convert the up- and down- ward crossing points of clipped speech into unipolar pulse train which is suitable to transmit in a band-limited channel. The noise suppression in pause may be achieved by means of slicing speech waves at a certain off-zero level that is higher than the noise amplitude. The articulation score of sliced speech at various levels, Fig. 9, shows that -30 to -40 db (db below the average peak amplitude) are optimum for the slice level in order to keep the sufficient intelligibility and to avoid the noise. The high frequency bias has a similar effect to this off-zero slicing in a sense and helps the function of slicer. The ambiguity of crossing points caused by the following reason makes another troublesome source of noise in the reproducing process from unipolar pulse train. When a maximum or minimum part of the wave just reaches the slicer level, the output of normal slicer has a form of imperfect or very narrow rectangular wave, from which such suitable position pulses being able to transmit through a band-limited channel could not be acquired. To overcome this effect, we use a slicer of modified mono-stable multivibrator having a time constant of τ, and then the speech wave is converted into such a rectangular wave train that the minimum duration between flip-over is limited to τ. It should be noted that the maximum τ without the deterioration of intelligibility is approximately 0. 18ms. which corresponds to the third formant frequency of speech sound. In this paper we discussed the above mentioned considerations of sliced speech and their applications for various voice communication systems.