RAV-Net Audio-Visual Speech Separation Samples

Each example presents the mixture, clean sources, and round-wise separation results. Audio files are sampled at 16 kHz.

Example 1 — Unseen-unheard test set

Sample ID: U-001; duration: 2.55 s; condition: Vox-S.

Mixture

Spectrogram of mixture

Speaker A

Clean source

Clean source A

Round 1

Round 1 output A

Round 2

Round 2 output A

Round 3

Round 3 output A

Round 4

Round 4 output A

Speaker B