Direct comparison
The panel stays fixed.
Every source clip, noise seed, normalization rule, and frozen 10-codebook codec matches the prior page. Enhancement now uses the sampler-benchmark winner, Heun-2: one predictor/corrector interval and exactly two model evaluations. “Lineage seen” means present in the step-105k warm-start training set. “Prior unseen” means disjoint from that set by path, filename, and content hash. This is not a strictly recording-disjoint DAC20k test: seven clips in each cohort share a recording with DAC20k train, and one lineage-seen clip is an exact DAC20k train path.
Ground truthNormalized source
NoisyCorrupt → frozen codec
EnhancedHeun-2 → frozen decoder
Loading the evaluation panel…