Duration Control With Variable Length Synthesis

via “duration control with variable-length synthesis”

Latent diffusion model for generating music and sound effects from text.

Unique: Implements duration control through temporal conditioning in the diffusion model rather than post-processing or concatenation, enabling seamless variable-length generation without artifacts. The model learns to scale temporal structure based on requested duration during training.

vs others: More flexible than fixed-length generators (which produce only 30-second or 60-second audio) because duration is user-controllable, and higher quality than concatenation-based approaches because the full audio is generated coherently in a single pass.