AudioGen · medium

Generate everyday sounds and environmental audio from descriptions.

GenerationOpen-source code1.5B
Original model card
Start here
Describe a recognizable sound event and its setting. Check whether the requested events actually happen, not only whether the texture sounds plausible.
Training data
AudioSet subset, BBC sound effects, AudioCaps, Clotho v2, VGG-Sound, FSD50K, Free To Use Sounds, Sonniss Game Effects, WeSoundEffects and Paramount Motion's Odeon collection.
Training rights
The model card lists research datasets and sound libraries with different terms; the weight license does not redistribute the original audio.
License & output terms
AudioCraft code: MIT. Weights: CC BY-NC 4.0.
Limitations
Research-oriented sound effects, not a dedicated full-song generator; limited long-range event control.
Technique
Autoregressive audio tokens
Runtime
Local AudioCraft inference; 1.5B parameter medium checkpoint.
Parameters
1.5B
Checkpoint (opens in a new tab)Model documentation (opens in a new tab)

Sources reviewed . Training and rights notes summarize provider disclosures. A model license does not clear every source recording or generated output. Parameter counts refer to the named variant or component, not total runtime memory.