Systems
- Baseline: The accent conversion system used in [1], available at https://github.com/guanlongzhao/ppg-gmm
- Proposed: The proposed accent conversion system, available at https://github.com/guanlongzhao/fac-via-ppg
Notes
- The L1 speakers speak the General American Accent
- YKWK's native language is Korean
- ZHAA's native language is Arabic
- Dataset (L2-ARCTIC corpus) [2]: https://psi.engr.tamu.edu/l2-arctic-corpus
- L2-ARCTIC recordings after noise removal (training data used in this study): Download link
- Please view this page in Google Chrome or Microsoft Edge for best quality
Compare baseline and proposed system
| L2 speaker | L1 reference speech | L2 reference speech | Baseline | Proposed |
|---|---|---|---|---|
| YKWK | ||||
| YKWK | ||||
| YKWK | ||||
| ZHAA | ||||
| ZHAA | ||||
| ZHAA |
Samples for mispronunciation and unexpected intonation by the proposed system
| Condition | L1 reference speech | L2 reference speech | Proposed |
|---|---|---|---|
| Mispronunciation | |||
| Unexpected intonation |
References
[1] G. Zhao, S. Sonsaat, J. Levis, E. Chukharev-Hudilainen, and R. Gutierrez-Osuna, "Accent conversion using phonetic posteriorgrams," in Proc. ICASSP, 2018.
[2] G. Zhao, S. Sonsaat, A. Silpachai, I. Lucic, E. Chukharev-Hudilainen, J. Levis, and R. Gutierrez-Osuna, "L2-ARCTIC: A non-native English speech corpus," in Proc. Interspeech, 2018.