Exploring synthetic data for cross-speaker style transfer in style representation based TTS
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916411559903232 |
|---|---|
| author | Ueda, Lucas H. Marques, Leonardo B. de M. M. Simões, Flávio O. Neto, Mário U. Runstein, Fernando Bó, Bianca Dal Costa, Paula D. P. |
| author_facet | Ueda, Lucas H. Marques, Leonardo B. de M. M. Simões, Flávio O. Neto, Mário U. Runstein, Fernando Bó, Bianca Dal Costa, Paula D. P. |
| contents | Incorporating cross-speaker style transfer in text-to-speech (TTS) models is challenging due to the need to disentangle speaker and style information in audio. In low-resource expressive data scenarios, voice conversion (VC) can generate expressive speech for target speakers, which can then be used to train the TTS model. However, the quality and style transfer ability of the VC model are crucial for the overall TTS model quality. In this work, we explore the use of synthetic data generated by a VC model to assist the TTS model in cross-speaker style transfer tasks. Additionally, we employ pre-training of the style encoder using timbre perturbation and prototypical angular loss to mitigate speaker leakage. Our results show that using VC synthetic data can improve the naturalness and speaker similarity of TTS in cross-speaker scenarios. Furthermore, we extend this approach to a cross-language scenario, enhancing accent transfer. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2409_17364 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Exploring synthetic data for cross-speaker style transfer in style representation based TTS Ueda, Lucas H. Marques, Leonardo B. de M. M. Simões, Flávio O. Neto, Mário U. Runstein, Fernando Bó, Bianca Dal Costa, Paula D. P. Audio and Speech Processing Sound Incorporating cross-speaker style transfer in text-to-speech (TTS) models is challenging due to the need to disentangle speaker and style information in audio. In low-resource expressive data scenarios, voice conversion (VC) can generate expressive speech for target speakers, which can then be used to train the TTS model. However, the quality and style transfer ability of the VC model are crucial for the overall TTS model quality. In this work, we explore the use of synthetic data generated by a VC model to assist the TTS model in cross-speaker style transfer tasks. Additionally, we employ pre-training of the style encoder using timbre perturbation and prototypical angular loss to mitigate speaker leakage. Our results show that using VC synthetic data can improve the naturalness and speaker similarity of TTS in cross-speaker scenarios. Furthermore, we extend this approach to a cross-language scenario, enhancing accent transfer. |
| title | Exploring synthetic data for cross-speaker style transfer in style representation based TTS |
| topic | Audio and Speech Processing Sound |
| url | https://arxiv.org/abs/2409.17364 |