Abstract

Keywords:

Method

Point-and-Play architecture diagram
Fig. 2 — Point-and-Play architecture. A CLAP encoder fine-tuned with LoRA encodes the mixture, while a frozen CLAP encoder encodes the source cue. Cue-Gated Statistics Pooling combines both representations into a timbre code of the cued source in the mixture and uses it to condition the note synthesizer. During training, the isolated source of the mixture provides supervision; it is not required at inference time.

Main Results

System and conditioning input CLAP ↑ FAD ↓ MSS ↓

Listen: Audio Samples

All systems below use the same mixture, four-note scale, and best-of-8 candidate selection. Every audio card is labeled with its own CLAP score against the ground-truth target, using the same independent judge as the Main Results table above. All audio on this page is 16 kHz mono WAV, so the effective bandwidth is capped at 8 kHz.

Citation

Full venue details and a BibTeX entry will be added once the paper's status is finalized.