ICASSP2010 Signals
This room is a regular office room (T60=520ms) with PC noise. The SNR of this recording is low due to the PC fan noise and air-conditioning. The recordings contain a total of 8 channels and reflect two different array configurations: Channels 1-5 are a cross-array with M=5 microphones and element-spacing of approx. d=4cm. The middle microphone is on channel 2. Channels 6-8 are an equiangular (60°) array with element-spacing of d=4cm. For reference we give the positions of all microphones in Matlab notation:
cross-array channels 1-5: mic1pos=[-d;0]; mic2pos=[0;0]; mic3pos=[d;0]; mic4pos=[0;-d]; mic5pos=[0;d]; equiangular array 6-8: mic6pos=[-(sqrt(3)/2) * d;0]; mic7pos=[0;-d/2]; mic8pos=[0;d/2];
To generate the mixtures used in the paper use the following MATLAB script: generate_mixtures_ICASSP.m
- room2_speech1_48kHz.wav
- room2_speech2_48kHz.wav
- room2_speech3_48kHz.wav
- room2_speech4_48kHz.wav
- room2_speech5_48kHz.wav
- room2_speech6_48kHz.wav
Blind Source Separation - Test Signals
Purpose
These recordings have been made in different environments (anechoic room, regular office environment with background noise such as air-conditioning and PC noise) with real human speakers and with loudspeakers, in order to test blind source separation algorithms. We invite other researchers to evaluate their algorithms on this data and would be happy to receive feedback on performance. The recordings represent rather difficult conditions with low SNR and/or high reverberation.
Filenaming Convention
All sources have been recorded one-by-one and saved as a multichannel WAV file. You can generate mixtures of signals by adding together two or more of the provided WAV files. The naming convention is as follows:
room<R>_source<S>_<F>.wav
where
R enumerates which room (see below).
S enumerates which source
F denotes the sampling rate (8kHz or 48kHz)
Room 1: Anechoic - Real Speakers
This room is an anechoic chamber and hence should be almost multipath free. The SNR of this recording is rather low because the real speakers spoke rather soft. Array configuration is a cross-array with M=5 microphones and element-spacing of approx. 4cm. The middle microphone is on channel 2.
- room1_speech1_8kHz.wav room1_speech1_48kHz.wav
- room1_speech2_8kHz.wav room1_speech2_48kHz.wav
- room1_speech3_8kHz.wav room1_speech3_48kHz.wav
- room1_speech4_8kHz.wav room1_speech4_48kHz.wav
- room1_speech5_8kHz.wav room1_speech5_48kHz.wav
- room1_speech6_8kHz.wav room1_speech6_48kHz.wav
Room 2: Office - Loudspeakers
This room is a regular office room (T60=500ms) with PC noise. The SNR of this recording is low due to the PC fan noise and air-conditioning. The recordings contain a total of 8 channels and reflect two different array configurations: Channels 1-5 are a cross-array with M=5 microphones and element-spacing of approx. d=4cm. The middle microphone is on channel 2. Channels 6-8 are an equiangular (60�) array with element-spacing of d=4cm. For reference we give the positions of all microphones in Matlab notation:
[...]