5 - Speech analysis

Summary

Instrumental analysis is necessary to understand how vocal communication works. Auditory transcriptions of speech can never isolate the acoustic cues that specify the sounds of speech. We can, for example, listen to as many carefully transcribed tokens of the syllables [di] and [du] as we care to, without ever discovering that different acoustic cues specify the “identical” sound [d] in these two syllables. As listeners, we have no more direct knowledge of the process of speech perception than we, as eaters of ice-cream, have of the enzyme systems that are involved in the digestion of sugar. If we want to study the digestion of sugar we have to make use of instrumental techniques. Likewise we have to make use of instrumental techniques for the analysis of speech.

In this chapter we will discuss both recent computer-implemented techniques and the sound spectrograph. The sound spectrograph was the instrument of choice for the analysis of speech from 1940 through the 1970s when computer-implemented methods began to be substituted. Computer-implemented techniques are now usually more accurate and can derive data that the sound spectrograph inherently cannot. However, the distinction between computer-implemented procedures and the sound spectrograph is beginning to blur; recent “digital” versions of the sound spectrograph are really dedicated microprocessors. Nevertheless, the sound spectrograph in either its traditional analog version or recent digital versions is often better suited for certain tasks, for example, showing formant frequency transitions, monitoring the speech samples of subjects, and tracking the acoustic changes consequent to rapid articulatory movements during physiological experiments, etc.