Intra-speaker phonetic variation in read speech: comparison with inter-speaker variability in a controlled population

Our knowledge of speech is historically built on data averaged across speakers or comparing different speakers. We therefore know little about the variability of speech produced by the same speaker: to what extent does it vary from one repetition to another and in what dimensions? In this study, we document the stability of speech and voice characteristics in 9 French speakers, on the reading of two texts recorded over ten sessions on different days over a two-month period. 21 features related to temporal, spectral, f0 and harmonicity aspects as well as their modulation between consecutive chunks are studied. The stability of these features between sessions is evaluated in comparison with their variability between speakers. Results show that short-term variability of energy in the 0-1kHz band, mean F0 and the slope of the LTAS vary the most between sessions for a given speaker, and are also among the speech and voice features that vary the most between speakers in this small cohort, while modulation features between consecutive chunks remain more stable across sessions.