Elamur Research⌕
☰
home / papers / 2026 / Automatic Prediction of Prominence and Boundary Strength from Text

Automatic Prediction of Prominence and Boundary Strength from Text

Summary

Text-to-speech systems struggle to infer prosody from text alone, though earlier work showed that oracle prosodic labels improve the rendering and controllability of synthetic speech. This paper asks whether continuous prominence and boundary strength values can be estimated from French text only, with the values obtained using the wavelet-based prosody tool developed by Suni et al. (2017). The authors compare three syllable-level approaches on nine speakers from the Mufasa audiobook corpus. The first is a FastSpeech2 variant with two extra predictors placed before the duration predictor and trained jointly with the TTS model. The second is BIPP, a standalone transformer whose decoder uses encoded phonetized syllables to query frozen camemBERT embeddings via cross-attention. The third is an auto-regressive BIPP that conditions on previously predicted prosodic values instead of syllables. Standard BIPP performed best. This suggests that pairing syllable information with semantic embeddings from a BERT-like model is the most effective route. The approach avoids expert ToBI-style annotation and reference audio, which opens the way to interpretable prosody control and style transfer.

Key points

  • Targets are the toolkit's continuous prominence and boundary values rather than its discretized 0–2 labels. They are computed per syllable instead of per word, from F0, energy and duration via a continuous wavelet transform.
  • Data: a subset of Mufasa (open-source French audiobooks) with 9 speakers (5 female, 4 male). The subset is unbalanced, with two female speakers accounting for almost a third of the audio.
  • BIPP uses a frozen camemBERT encoder. Its last-layer embedding outperformed max-pooling over all hidden layers. The model also has a convolutional syllable encoder (output dimension 256), a 2-head transformer decoder, speaker embeddings and parallel prominence and boundary heads.
  • Cross-attention removes the need to explicitly align syllables with camemBERT's subword tokens.
  • Salient events were analyzed separately, using thresholds at 80% (prominence) and 90% (boundaries) of the cumulative distribution. Planned future work includes integrating BIPP into TTS, testing style transfer and building GUI-based prosody control.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/3k3ii2w38tnj

@inproceedings{mas2026automatic,
  title = {{Automatic Prediction of Prominence and Boundary Strength from Text}},
  author = {Mas, Pauline and Vythelingum, Kévin and Chevelu, Jonathan and Ouédraogo, Marion and Lolive, Damien and Rosec, Olivier},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/3k3ii2w38tnj},
  url = {https://doi.org/10.63317/3k3ii2w38tnj}
}

References

22 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Chung-Ming Chien and Hung-yi Lee. 2021. Hierarchical prosody modeling for non-autoregressive speech synthesis. In 2021 IEEE Spoken Language Technology Workshop (SLT), pages 446– 453. find
  2. Ziqian Dai, Jianwei Yu, Yan Wang, Nuo Chen, Yanyao Bian, Guangzhi Li, Deng Cai, and Dong Yu. 2022. Automatic Prosody Annotation with Pre-Trained Text-Speech Model. In Interspeech 2022, pages 5513–5517. find
  3. Wietse De Vries, Andreas Van Cranenburgh, and Malvina Nissim. 2020. What’s so special about BERT’s layers? A closer look at the NLP pipeline in monolingual and multilingual models. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 4339–4350, Online. Association for Comp… find
  4. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,… find
  5. Esther Grabe, Francis Nolan, and Kimberley J. Farrar. 1998. Ivie - a comparative transcription system for intonational variation in english. In 5th International Conference on Spoken Language Processing (ICSLP 1998), page paper 0099. find
  6. Ganesh Jawahar, Benoît Sagot, and Djamé Seddah. 2019. What does BERT learn about the structure of language? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3651–3657, Florence, Italy. Association for Computational Linguistics. find our card
  7. Sofoklis Kakouros and Johannah O’Mahony. 2023. What does BERT learn about prosody? find
  8. Taku Kudo and John Richardson. 2018. SentencePiece: A simple and language independent subword tokenizer and detokenizer for neural text processing. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 66– 71, Brussels, Belgium. Assoc… find our card
  9. Yongjie Lin, Yi Chern Tan, and Robert Frank. 2019. Open Sesame: Getting inside BERT’s Linguistic Knowledge. In Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 241–253, Florence, Italy. Association for Computational Linguistics. find our card
  10. Louis Martin, Benjamin Muller, Pedro Javier Ortiz Suárez, Yoann Dupont, Laurent Romary, Éric de la Clergerie, Djamé Seddah, and Benoît Sagot. 2020. CamemBERT: a tasty French language model. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7203–7219, O… find our card
  11. Hosein Mohebbi, Ali Modarressi, and Mohammad Taher Pilehvar. 2021. Exploring the Role of BERT Token Representations to Explain Sentence Probing Results. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 792–806, Online and Punta Cana, Dominican Republic… find
  12. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2022. Robust speech recognition via large-scale weak supervision. find our card
  13. Yi Ren, Chenxu Hu, Xu Tan, Tao Qin, Sheng Zhao, Zhou Zhao, and Tie-Yan Liu. 2021. FastSpeech 2: Fast and high-quality end-to-end text to speech. In International Conference on Learning Representations. find
  14. Noam Shazeer. 2019. Fast transformer decoding: One write-head is all you need. find
  15. Atli Thor Sigurgeirsson and Simon King. 2023. Do prosody transfer models transfer prosody? In ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. find
  16. Kim Silverman, Mary Beckman, John Pitrelli, Mori Ostendorf, Colin Wightman, Patti Price, Janet Pierrehumbert, and Julia Hirschberg. 1992. Tobi: a standard for labeling english prosody. In 2nd International Conference on Spoken Language Processing (ICSLP 1992), pages 867–870. find
  17. Antti Suni, Sofoklis Kakouros, Martti Vainio, and Juraj Šimko. 2020. Prosodic Prominence and Boundaries in Sequence-to-Sequence Speech Synthesis. In Speech Prosody 2020, pages 940– 944, ISCA. ISCA. find
  18. Antti Suni, Juraj Šimko, Daniel Aalto, and Martti Vainio. 2017. Hierarchical representation and estimation of prosody using continuous wavelet transform. Computer Speech & Language, 45:123–136. find
  19. Jacqueline Vaissière. 2020. Chapitre IX - Prosodie. In La Phonétique, volume 4e éd. of Que sais-je ?, pages 96–117. Presses Universitaires de France, Paris cedex 14. find
  20. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Proceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, page 6000–6010, Red Hook, NY, USA. Cur… find
  21. Yuxiang Zou, Shichao Liu, Xiang Yin, Haopeng Lin, Chunfeng Wang, Haoyu Zhang, and Zejun Ma. 2021. Fine-grained prosody modeling in neural speech synthesis using tobi representation. In Interspeech 2021, pages 3146–3150. 9. find
  22. Language Resource References Sini, Aghilas and Wadoux, Lily and Perquin, Antoine and Vidal, Gaëlle and Guennec, David and Lolive, Damien and Alain, Pierre and Barbot, Nelly and Chevelu, Jonathan and Delhay, Arnaud. 2022. Techniques de synthèse vocale neuronale à l’épreuve des données d’apprentissage… find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mas-2026-automatic-prediction-prominence-boundary-strength/