Reliability of clinical voice parameters captured with smartphones – measurements of added noise and spectral tilt
View/ Open
Date
2019-09-20Author
Schaeffler, Felix
Jannetts, Stephen
Beck, Janet M.
Metadata
Show full item recordCitation
Schaeffler, F., Jannetts, S. & Beck, J. M. (2019) Reliability of clinical voice parameters captured with smartphones – measurements of added noise and spectral tilt. In: Proceedings of the 20th Annual Conference of the International Speech Communication Association INTERSPEECH, Graz, Austria, 15-19 September 2019, pp. 2523-2527.
Abstract
Smartphones have become powerful tools for data capture due
to their computational power, internet connectivity, high
quality sensors and user-friendly interfaces. This also makes
them attractive for the recording of voice data that can be
analysed for clinical or other voice health purposes. This
however requires detailed assessment of the reliability of
voice parameters extracted from smartphone recordings. In a
previous study we analysed reliability of measures of
periodicity and periodicity deviation, with very mixed results
across parameters. In the present study we extended this
analysis to measures of added noise and spectral tilt. We
analysed systematic and random error for six frequently used
acoustic parameters in clinical acoustic voice quality analysis.
22 speakers recorded sustained [a] and a short passage with a
studio microphone and four popular smartphones
simultaneously. Acoustic parameters were extracted with Praat
and smartphone recordings were compared to the studio
microphone. Results indicate a small systematic error for
almost all parameters and smartphones. Random errors
differed substantially between parameters. Our results suggest
that extraction of acoustic voice parameters with mobile
phones is not without problems and different parameters show
substantial differences in reliability. Careful individual
assessment of parameters is therefore recommended before
use in practice.