The sounds of speech are generated by the speech mechanism of a vast variety of different speakers, even if we confine our attention to the sounds of one language. All these speakers have a larynx and vocal tract of different dimensions, they have a different upbringing and different speech habits, a different emotional make-up and voice quality. Even one individual speaker talking at different times will produce sounds which are acoustically very different so that any utterance we may take for purposes of acoustic analysis can be only one sample out of millions of possible ones. Some impression of the extreme variability to be met with in utterances from any considerable number of different speakers can be gained from the plots in Fig. 42. These give the frequency of F1 and F2 for ten different vowels in utterances by 76 speakers of American English. Even here the variation is very much reduced by the fact that the analyses are of words spoken in isolation and not in running speech. It is clear that if we are going to be concerned with the acoustic analysis of speech sounds we must be prepared to encounter this degree of variability.

Because language systems work, however, despite acoustic variability, it is both useful and meaningful to make generalizations about the acoustic characteristics of the different classes of sound that make up a linguistic sound system.