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

TL;DR

This paper systematically explores the design space of text-to-audio models, introducing a large synthetic dataset, comparing various model choices, and proposing ETTA, which outperforms baselines and excels in creative audio generation.

Contribution

It provides a comprehensive empirical analysis of TTA model components and introduces ETTA, a new model that advances quality and creative capabilities.

Findings

ETTA outperforms baseline models on AudioCaps and MusicCaps.

Sampling strategies significantly impact quality and speed trade-offs.

Synthetic dataset AF-Synthetic enhances training and evaluation.

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.