TY - GEN
T1 - Variation in spectral slope and interharmonic noise in Cantonese tones
AU - Rose, Phil
N1 - Publisher Copyright:
Copyright © 2020 ISCA
PY - 2020
Y1 - 2020
N2 - To provide reference data for studies of voice quality variation in lexical tone, an experiment is described to investigate the nature of intrinsic variation in spectral slope and interharmonic noise for Cantonese citation tones. 23 spectral slope and interharmonic noise measures are extracted with VoiceSauce from the tones on /o/ Rhymes of five male and five female speakers of conservative Cantonese. Significant correlation between F0 and both spectral slope and interharmonic noise is demonstrated. It is shown with probabilistic bivariate discriminant analysis that even tones with no extrinsic voice quality differences can be identified at rates considerably above chance from a combination of their spectral slope and interharmonic noise. Male tones, with a minimal error rate of 5.7%, are identified twice as well as female, with a minimal error rate of 14.5%. Combinations with uncorrected spectral slopes perform better than corrected. The best combinations for both sexes involve slope parameters H2H4 (difference between the 4th and 2nd harmonic amplitudes); and H42K (difference between the 4th harmonic and nearest harmonic to 2 kHz), irrespective of noise parameters. The worst combinations involve CPP (cepstral peak prominence) as a noise parameter.
AB - To provide reference data for studies of voice quality variation in lexical tone, an experiment is described to investigate the nature of intrinsic variation in spectral slope and interharmonic noise for Cantonese citation tones. 23 spectral slope and interharmonic noise measures are extracted with VoiceSauce from the tones on /o/ Rhymes of five male and five female speakers of conservative Cantonese. Significant correlation between F0 and both spectral slope and interharmonic noise is demonstrated. It is shown with probabilistic bivariate discriminant analysis that even tones with no extrinsic voice quality differences can be identified at rates considerably above chance from a combination of their spectral slope and interharmonic noise. Male tones, with a minimal error rate of 5.7%, are identified twice as well as female, with a minimal error rate of 14.5%. Combinations with uncorrected spectral slopes perform better than corrected. The best combinations for both sexes involve slope parameters H2H4 (difference between the 4th and 2nd harmonic amplitudes); and H42K (difference between the 4th harmonic and nearest harmonic to 2 kHz), irrespective of noise parameters. The worst combinations involve CPP (cepstral peak prominence) as a noise parameter.
KW - Cantonese citation tones
KW - HNR-ratio
KW - Phonation type
KW - Spectral slope
KW - Voice quality
UR - http://www.scopus.com/inward/record.url?scp=85098219700&partnerID=8YFLogxK
U2 - 10.21437/Interspeech.2020-1954
DO - 10.21437/Interspeech.2020-1954
M3 - Conference contribution
SN - 9781713820697
T3 - Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
SP - 1933
EP - 1937
BT - Interspeech 2020
PB - International Speech Communication Association
T2 - 21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020
Y2 - 25 October 2020 through 29 October 2020
ER -