ETTA: Elucidating the Design Space of Text-to-Audio Models

Sang-Gil Lee, Zhifeng Kong, Arushi Goel, Sungwon Kim, Rafael Valle, Bryan Catanzaro
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:33157-33180, 2025.

Abstract

Recent years have seen significant progress in Text-To-Audio (TTA) synthesis, enabling users to enrich their creative workflows with synthetic audio generated from natural language prompts. Despite this progress, the effects of data, model architecture, training objective functions, and sampling strategies on target benchmarks are not well understood. With the purpose of providing a holistic understanding of the design space of TTA models, we set up a large-scale empirical experiment focused on diffusion and flow matching models. Our contributions include: 1) AF-Synthetic, a large dataset of high quality synthetic captions obtained from an audio understanding model; 2) a systematic comparison of different architectural, training, and inference design choices for TTA models; 3) an analysis of sampling methods and their Pareto curves with respect to generation quality and inference speed. We leverage the knowledge obtained from this extensive analysis to propose our best model dubbed Elucidated Text-To-Audio (ETTA). When evaluated on AudioCaps and MusicCaps, ETTA provides improvements over the baselines trained on publicly available data, while being competitive with models trained on proprietary data. Finally, we show ETTA’s improved ability to generate creative audio following complex and imaginative captions – a task that is more challenging than current benchmarks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-lee25m, title = {{ETTA}: Elucidating the Design Space of Text-to-Audio Models}, author = {Lee, Sang-Gil and Kong, Zhifeng and Goel, Arushi and Kim, Sungwon and Valle, Rafael and Catanzaro, Bryan}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {33157--33180}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/lee25m/lee25m.pdf}, url = {https://proceedings.mlr.press/v267/lee25m.html}, abstract = {Recent years have seen significant progress in Text-To-Audio (TTA) synthesis, enabling users to enrich their creative workflows with synthetic audio generated from natural language prompts. Despite this progress, the effects of data, model architecture, training objective functions, and sampling strategies on target benchmarks are not well understood. With the purpose of providing a holistic understanding of the design space of TTA models, we set up a large-scale empirical experiment focused on diffusion and flow matching models. Our contributions include: 1) AF-Synthetic, a large dataset of high quality synthetic captions obtained from an audio understanding model; 2) a systematic comparison of different architectural, training, and inference design choices for TTA models; 3) an analysis of sampling methods and their Pareto curves with respect to generation quality and inference speed. We leverage the knowledge obtained from this extensive analysis to propose our best model dubbed Elucidated Text-To-Audio (ETTA). When evaluated on AudioCaps and MusicCaps, ETTA provides improvements over the baselines trained on publicly available data, while being competitive with models trained on proprietary data. Finally, we show ETTA’s improved ability to generate creative audio following complex and imaginative captions – a task that is more challenging than current benchmarks.} }
Endnote
%0 Conference Paper %T ETTA: Elucidating the Design Space of Text-to-Audio Models %A Sang-Gil Lee %A Zhifeng Kong %A Arushi Goel %A Sungwon Kim %A Rafael Valle %A Bryan Catanzaro %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-lee25m %I PMLR %P 33157--33180 %U https://proceedings.mlr.press/v267/lee25m.html %V 267 %X Recent years have seen significant progress in Text-To-Audio (TTA) synthesis, enabling users to enrich their creative workflows with synthetic audio generated from natural language prompts. Despite this progress, the effects of data, model architecture, training objective functions, and sampling strategies on target benchmarks are not well understood. With the purpose of providing a holistic understanding of the design space of TTA models, we set up a large-scale empirical experiment focused on diffusion and flow matching models. Our contributions include: 1) AF-Synthetic, a large dataset of high quality synthetic captions obtained from an audio understanding model; 2) a systematic comparison of different architectural, training, and inference design choices for TTA models; 3) an analysis of sampling methods and their Pareto curves with respect to generation quality and inference speed. We leverage the knowledge obtained from this extensive analysis to propose our best model dubbed Elucidated Text-To-Audio (ETTA). When evaluated on AudioCaps and MusicCaps, ETTA provides improvements over the baselines trained on publicly available data, while being competitive with models trained on proprietary data. Finally, we show ETTA’s improved ability to generate creative audio following complex and imaginative captions – a task that is more challenging than current benchmarks.
APA
Lee, S., Kong, Z., Goel, A., Kim, S., Valle, R. & Catanzaro, B.. (2025). ETTA: Elucidating the Design Space of Text-to-Audio Models. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:33157-33180 Available from https://proceedings.mlr.press/v267/lee25m.html.

Related Material