The consistency of formant measurements in high quality audio data: the effect of agreeing measurement procedures

This study examines the inter-analyst robustness of formant frequency (F) measurements before and after agreeing a common measurement procedure. Three laboratories used the latest available version of the same formant extraction software to extract F1, F2 and F3 values from the same sets of high quality audio recordings of six repetitions of read utterances containing six target monophthongs, /iː, æ, ɑː, ɔː, ʊ, uː/. The subjects were two sets of twenty British males with a similar accent profile. Results are presented which show the extent to which analyst, methodology, vowel, and speaker affect the consistency of the values extracted. Agreeing the methodology reduced the extent of variation among the results from the three laboratories as might be expected. However, there remained some differences and even occasionally a greater disparity among the measurements. The paper suggests sources of these differences and makes proposals for a consistent logging of settings and times used when extracting F values. Further, it recommends that practitioners aim to establish an agreed protocol for forensic analysis of formants to facilitate the validation of evidence.