Título: “Speech rate estimation: how long should the utterance be?”

Pablo Arantes

Resumo

In this study we try to develop a methodology to estimate the minimum
duration a speech should have in order to the speech rate (SR)
derived from it can be said to be representative. Change point
analysis was applied to a set of time series comprised of cumulative
SR calculated throughout samples of read speech produced by four
speakers of Brazilian Portuguese in three rates (slow, normal and
fast) to determine how long it takes for SR estimates to stabilize
around a narrow range of values. SR was defined here as the number of
intervals between consecutive vowel onsets (V-V units) per unit of
time. Results show that SR values stabilize between 5 and 12 seconds,
a time interval that contains between 36 and 68 V-V units. Also, SR in
fast rate tends to stabilize sooner.