Sound Effect Variation Generation

via “text-to-sound effect generation”

Meta's library for music and audio generation.

Unique: Reuses MusicGen's architecture but with domain-specific training on sound effect datasets and adapted conditioning systems; enables the same efficient token-based generation pipeline for non-musical audio without separate model implementations.

vs others: More flexible than sample-based sound libraries and faster than real-time synthesis engines; open-source implementation allows fine-tuning on custom sound datasets.