This page provides the sound files used in the article "Underdetermined source separation with structured source priors", submitted for publication at ICA'04.
The structure of the table below is the same as the structure of Table 1 in the article.
Mixture: x
Sources: s1 (cello), s2 (clarinet), s3 (violin)
| Mixing matrix: A= | 1.64 | 0.33 | 0.92 |
| 0.95 | 0.56 | 0.92 |
| Cues | Method | Extracted sources | SIR (dB) | SAR (dB) | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Spatial diversity | All sources | s1 | s2 | s3 | 10 | 8 | -5 | +inf | +inf | +inf |
| Closest source | s1 | s2 | s3 | 36 | 26 | 18 | 11.6 | 10.3 | 5.6 | |
| 1 or 2 closest sources | s1 | s2 | s3 | 27 | 25 | 15 | 13.8 | 13.9 | 5.9 | |
| Source priors | Bernoulli state priors | s1 | s2 | s3 | 13 | 12 | 6 | 11.8 | 10.6 | -3.0 |
| Spatial diversity + Source priors | Bernoulli state priors | s1 | s2 | s3 | 23 | 22 | 34 | 17.1 | 16.8 | 7.0 |
| Markov state priors | s1 | s2 | s3 | 30 | 31 | 23 | 17.2 | 16.8 | 8.4 | |
| Oracle state sequence | s1 | s2 | s3 | 31 | 35 | 23 | 18.7 | 18.6 | 10.5 | |
| Oracle | Oracle | s1 | s2 | s3 | 49 | 49 | 44 | 24.4 | 30.0 | 21.9 |