DialogueSidon2 — Two-Speaker Dialogue Separation & Restoration

Upload a degraded or noisy audio/video recording of a two-speaker conversation. DialogueSidon2 separates the two speakers and restores clean 24 kHz speech from the mixture in one forward pass, without sampling: the same input always gives the same output. Inputs up to 30 s are processed in one shot. Longer inputs are processed in 30 s windows with 5 s overlap crossfade and automatic speaker re-alignment across windows. Model: sarulab-speech/DialogueSidon2 · previous version: DialogueSidon