Acoustic Token Refinement For Perceptual Quality

via “fine-grained audio detail synthesis via non-causal refinement”

Open-source text-to-audio — speech, music, sound effects, 13+ languages, runs locally.

Unique: Uses non-causal (bidirectional) attention to refine audio tokens, allowing each position to condition on future context for higher-quality reconstruction than causal-only approaches

vs others: Bidirectional refinement produces more natural audio than single-pass causal models; hierarchical approach enables faster coarse generation with optional fine refinement

via “diffusion-based acoustic refinement with configurable denoising steps”

A high quality multi-voice text-to-speech library

Unique: Uses diffusion-based iterative denoising in mel spectrogram space rather than waveform space, making refinement computationally efficient while capturing acoustic details. Configurable step count enables explicit quality/speed tradeoff without model retraining.

vs others: More efficient than waveform-space diffusion (like DiffWave) because mel spectrograms are lower-dimensional; more flexible than fixed-quality systems because step count is tunable; captures acoustic details better than single-pass refinement networks.

A model by Google Research for generating high-fidelity music from text descriptions.