SR-FD VoxCPM2 Audio Demo

Side-by-side audio from the current VoxCPM2 low-NFE experiments: base, LoRA fine-tuning, SR-FD, CFS-FD, and MeanFlow controls.

Comparison Set

Source report

Prompt

Reference audio and generated samples use the same LibriTTS eval prompt id.

Aggregate Metrics

WER and CER are percentages copied from the source metrics files for the whole eval cell.

Audio Samples

Use the controls below to compare the selected prompt across systems.