Combining Masked Language Modeling and Cross-Modal Contrastive Learning for Prosody-Aware TTS
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866912996259790848 |
|---|---|
| author | Borodin, Kirill Kudryavtsev, Vasiliy Maslov, Maxim Vasiliev, Nikita Gorodnichev, Mikhail Mkrtchian, Grach |
| author_facet | Borodin, Kirill Kudryavtsev, Vasiliy Maslov, Maxim Vasiliev, Nikita Gorodnichev, Mikhail Mkrtchian, Grach |
| contents | We investigate multi-stage pretraining for prosody modeling in diffusion-based TTS. A speaker-conditioned dual-stream encoder is trained with masked language modeling followed by SigLIP-style cross-modal contrastive learning using mixed-phoneme batches, with an additional same-phoneme refinement stage studied separately. We evaluate intrinsic text-audio retrieval and downstream synthesis in Grad-TTS and a latent diffusion TTS system. The two-stage curriculum (MLM + mixed-phoneme contrastive learning) achieves the best overall synthesis quality in terms of intelligibility, speaker similarity, and perceptual measures. Although same-phoneme refinement improves prosodic retrieval, it reduces phoneme discrimination and degrades synthesis. These findings indicate that improvements in embedding-space metrics do not necessarily translate to better generative performance and highlight the need to balance phoneme discrimination and prosodic sensitivity in TTS pretraining. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_01247 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Combining Masked Language Modeling and Cross-Modal Contrastive Learning for Prosody-Aware TTS Borodin, Kirill Kudryavtsev, Vasiliy Maslov, Maxim Vasiliev, Nikita Gorodnichev, Mikhail Mkrtchian, Grach Sound Audio and Speech Processing We investigate multi-stage pretraining for prosody modeling in diffusion-based TTS. A speaker-conditioned dual-stream encoder is trained with masked language modeling followed by SigLIP-style cross-modal contrastive learning using mixed-phoneme batches, with an additional same-phoneme refinement stage studied separately. We evaluate intrinsic text-audio retrieval and downstream synthesis in Grad-TTS and a latent diffusion TTS system. The two-stage curriculum (MLM + mixed-phoneme contrastive learning) achieves the best overall synthesis quality in terms of intelligibility, speaker similarity, and perceptual measures. Although same-phoneme refinement improves prosodic retrieval, it reduces phoneme discrimination and degrades synthesis. These findings indicate that improvements in embedding-space metrics do not necessarily translate to better generative performance and highlight the need to balance phoneme discrimination and prosodic sensitivity in TTS pretraining. |
| title | Combining Masked Language Modeling and Cross-Modal Contrastive Learning for Prosody-Aware TTS |
| topic | Sound Audio and Speech Processing |
| url | https://arxiv.org/abs/2604.01247 |