| Project/Area Number |
19K13162
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 02060:Linguistics-related
|
|---|
| Research Institution | The University of Aizu |
|---|
Principal Investigator |
Perkins Jeremy 会津大学, コンピュータ理工学部, 上級准教授 (30725635)
|
|---|
| Project Period (FY) |
2019-04-01 – 2023-03-31
|
| Project Status |
Completed (Fiscal Year 2022)
|
| Budget Amount *help |
¥1,300,000 (Direct Cost: ¥1,000,000、Indirect Cost: ¥300,000)
Fiscal Year 2020: ¥520,000 (Direct Cost: ¥400,000、Indirect Cost: ¥120,000)
Fiscal Year 2019: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000)
|
|---|
| Keywords | Psychoacoustic roughness / glottalized consonants / spectral tilt / Thai / Korean / machine learning / random forest / roughness / f0 / laryngeal constriction / laryngealized consonants / creakiness / psychoacoustic roughness / electroglottography / EGG / creaky phonation / tone / phonation / consonants / psychacoustic roughness / glotallized consonants / coarticulation / creaky |
|---|
| Outline of Research at the Start |
This research investigates the use of a new method to identify consonant sounds as involving glottalization (throat constriction). This method, psychoacoustic roughness, is unique in that it is linked to the way humans perceive creaky sounds, making it more suitable than commonly used acoustic measures. Linguistic researchers doing phonetics field work can benefit from this work because it involves working directly with sound file, is not invasive and doesn't require any specialized equipment. Recordings of Thai and Korean speakers will provide the data for this study.
|
| Outline of Final Research Achievements |
Data was collected from Korean and Thai that showed psychoacoustic roughness can identify glottalized consonants in both Thai and Korean. However, this result was more robust in Thai than in Korean, where other methods yielded larger effects for Korean. Finally, machine learning methods were used, training a model on commonly used acoustic measurements. This method was shown to be a promising method for linguists to use in production studies with multiple acoustic measures. It confirmed previous studies on Korean that f0 is the primary acoustic measure that distinguishes aspirated and lenis consonants. It also showed that voice-onset-time is the primary measure distinguishing tense consonants from others, showing that glottalization is a secondary feature, and not necessary to distinguish tense consonants from lenis or aspirated ones. These results will be presented at a top-level international phonetics conference, with associated publications.
|
| Academic Significance and Societal Importance of the Research Achievements |
This research showed that acoustic measures that were traditionally used for voice quality in vowels can also identify glottal constriction in consonants (in particular, psychoacoustic roughness). Also, machine learning was used to assess which acoustic measures can distinguish groups of sounds.
|
