35 Managing Data Workflows for Untrained Forced Alignment: Examples from Costa Rica, Mexico, the Cook Islands, and Vanuatu

Forced alignment is a technique to align the audio signal of a spoken utterance with its transcription, so that the boundaries between words, and even between phones,1 can be determined automatically (Wightman & Talkin 1997). Figure 35.1 shows an example of this in Spanish, where the spectrogram of an utterance is accompanied by lines indicating the approximate temporal limits of words and individual phones within the recording. The main advantage of this technique is that it accelerates phonetic research: the algorithm can tag these boundaries approximately thirty times faster than expert humans can (Labov, Rosenfelder, & Fruehwald 2013).