SoundRadar: audio samples

Every source of each mixture is extracted, with one row per source: reference, output and direction estimate. Spectrograms show the first microphone in dB relative to the maximum of the mixture; the audio files keep the four microphones and the relative levels of the mixture. Direction figures: azimuth φ and colatitude θ of the ground truth (gray) and of the estimate (source color); dotted line = query time. The intermediate decoders (blocks 4 and 2) are folded under each source. Click a figure to enlarge.