Acoustic representations and segments

Acoustic representations are rarely static in the way that transcriptions are. In the waveform and spectrogram of ‘spend’ (Figure 3.3), many things change simultaneously: the amplitude of the signal and the formants in particular are not static. In speech, many articulations do not start and stop quite synchronously. Looking at the spectrogram we can identify six or seven more or less stable portions. On the other hand, this utterance is transcribed broadly as [spεnd], which implies five discrete units.

Transcriptions and acoustic representations capture different kinds of truth about speech. [spεnd] captures the fact that English speakers conceive of this word as having five distinct sounds. English speakers’ intuitions about how many segments there are do not match up with what our eyes might tell us. The acoustic representation captures the fact that in speaking, the articulators are rarely static. When articulators move, these movements have acoustic consequences, and this very fluidity helps to make everyday speech easier to perceive. Both the acoustic and written representations convey important but different information about speech.

Transcriptions may have a generality to them which acoustic representations do not. A broad transcription represents many of the important details of the speech of a whole community of speakers, which is why such transcriptions are used in dictionaries. On the other hand, acoustic representations capture details and facts about one utterance on one occasion by one speaker (as may an impressionistic transcription); if the speaker changed, or if the same speaker produced the same word e.g. more slowly, then many of the details of the acoustic representation would also change. So acoustic representations may be less useful from the point of view of representing facts about language.