This page contains some demonstrations of algorithms I've worked on. It is not very complete. In fact, it currently contains only one demo, which is not interactive. Nor is it multimedia: the only modality used is audio.

Contextual ICA

The first demo applies the Contextual ICA blind source separation algorithm. Lucas Parra and I digitally extracted ten five-second clips from ten audio CDs. These were digitally mixed, without time delays or echos, and with random gains, to form the output of a synthetic microphone. Ten such microphone outputs were synthesized.

These synthetic microphone outputs formed the input to the Bell-Sejnowski Independent Components Analysis algorithm. The sources are somewhat separated in the output of the Bell-Sejnowski ICA algorithm, but not fully.

The same synthetic microphone outputs were then used as input to our new cICA algorithm (see my publications page for technical details). The sources are almost fully separated in the output of cICA.

Fully audio demo

If you would like to hear a fully audio demonstration, the following ten tracks can be either played consecutively, or recorded on an analog tape or a CD-R. To ease the later course, the audio format chosen is Redbook compatible, so the disks can be played in any audio CD player.

Tracks 1-8 constitute the demo proper. Tracks 9 and 10 are bonus tracks, not on the original album!!!

track description length 16-bit 44.1 kHz stereo 16-bit 11.025 kHz mono
1 Narrator: Intro to contextual ica and description of demo 1:30 15,541 kb 1,945 kb
2 Narrator: ``Part 1: the outputs of the ten synthetic microphones'' 0:05 816 kb 103 kb
3 The outputs of the ten synthetic microphones 1:00 10,228 kb 1,280 kb
4 Narrator: ``Part 2: the ten sources recovered by the ICA algorithm'' 0:05 888 kb 112 kb
5 The ten sources recovered by the ICA algorithm 1:00 10,228 kb 1,280 kb
6 Narrator: ``Part 3: the ten sources recovered by contextual ICA'' 0:06 932 kb 118 kb
7 The ten sources recovered by contextual ICA 1:00 10,228 kb 1,280 kb
8 Narrator: Conclusion and future work 0:13 2,104 kb 264 kb
9 Narrator: ``The ten original sources'' 0:03 400 kb 52 kb
10 The ten original sources 1:00 10,228 kb 1,280 kb
total Tar file containing all tracks 6:02 38,465 kb 7,711 kb

Data for Source Separation Experiments

If you would like to experiment on source separation using the same data set, I am very pleased to make it available, for research purposes only! Please, not to be used for entertainment!

Although the demo used only ten audio sources, I have sixteen available for experiments. They are available in two forms, via the following pages: as .wav files, as compressed ascii text, and, for truly blind experiments, as ten pre-mixed .wav files, the same ones used in the demo above. There are also some notes describing exactly where these audio files came from.

In our experiments with cICA, we converted to mono and downsampled by a factor of five. For a truly fair comparison with another algorithm, you should do the same.

Barak Pearlmutter