14thAnnual Conference of the International Speech Communication Association

Lyon, France
August 25-29, 2013

Discrimination Between Fricative and Affricate in Japanese Using Time and Spectral Domain Variables

Kimiko Yamakawa, Shigeaki Amano

Aichi Shukutoku University, Japan

Previous studies on Japanese fricatives and affricates revealed that [s] and [ts] are classified by variables in the time domain (Yamakawa et al., 2012), whereas [ts] and [tC] as well as [s] and [tC] are classified by variables in the spectral domain (Yamakawa & Amano, 2011). To gain an integrated perspective on these findings, this study examined whether [s] [ts], and [tC] can be classified by a single discriminant model. Using variables in the time and spectral domains, canonical discriminant analysis was performed on word materials with [s], [ts], and [tC] pronounced by both single and multiple Japanese speakers. The time domain variables were a combination of the rise duration and sum of steady and decay durations of the three consonants. The spectral domain variable was their intensity obtained from a one-third-octave bandpass filter with a center frequency of 3150 Hz. The results showed that [s], [ts], and [tC] were successfully classified with high discriminant ratios of 92.6% and 90.2% for single and multi-speaker materials, respectively. This means that, in both cases, the single discriminant model can discriminate [s], [ts], and [tC] with the variables in the time and spectral domains.

Full Paper

Bibliographic reference.  Yamakawa, Kimiko / Amano, Shigeaki (2013): "Discrimination between fricative and affricate in Japanese using time and spectral domain variables", In INTERSPEECH-2013, 978-981.