This page presents the audio results of a number of source separation experiments carried
out in the Audio Lab
of the Department of
Electronics, University of York. The audio
accompanies the research work carried out during my PhD, which was supervised
by Dr John Szymanski and presented in detail in my PhD thesis
(see link on the right).
The dB values refer to the RMS energy difference between sources j and j + 1. For
example, a ''5 dB'' value indicates that the RMS energy of source 1 is 5 dB higher compared to
the RMS energy of source 2, 10 dB higher compared to the RMS energy of source 3, and so
on. For a detailed description and discussion concerning the particular experiments, the
proposed separation system and the alternative techniques it is
compared with [1, 2], please see Ch. 5 and Ch. 6 of the thesis.
For the purpose of carrying out comparisons, the WAV versions of the
audio files are also provided in an archived form.
References
[1] Z. Duan and B. Pardo. Soundprism: An online system for score-informed
source separation of music audio. IEEE Selected Topics in Signal
Processing (submitted), 2011.
[2] Y. Li, J. Woodruff, and D. L. Wang. Monaural musical sound separation
based on pitch and common amplitude modulation. IEEE Trans. Audio,
Speech, and Language Processing, 17(7):1361-1371, 2009.
melody_mix1, O dB 
|
|
source 1 |
source 2 |
source 3 |
residual |
| Original |
|
|
|
|
| Automatic |
|
|
|
|
| Disentangling |
|
|
|
|
| Use of residual |
|
|
|
|
| MIDI |
|
|
|
|
| Duan and Pardo |
|
|
|
|
| Li et al. |
|
|
|
|
melody_mix1, -5 dB 
|
|
source 1 |
source 2 |
source 3 |
residual |
| Original |
|
|
|
|
| Automatic |
|
|
|
|
| Disentangling |
|
|
|
|
| Use of residual |
|
|
|
|
| MIDI |
|
|
|
|
| Duan and Pardo |
|
|
|
|
| Li et al. |
|
|
|
|
melody_mix1, 5 dB 
|
|
source 1 |
source 2 |
source 3 |
residual |
| Original |
|
|
|
|
| Automatic |
|
|
|
|
| Disentangling |
|
|
|
|
| Use of residual |
|
|
|
|
| MIDI |
|
|
|
|
| Duan and Pardo |
|
|
|
|
| Li et al. |
|
|
|
|
melody_mix2, O dB 
|
|
source 1 |
source 2 |
source 3 |
residual |
| Original |
|
|
|
|
| Automatic |
|
|
|
|
| Disentangling |
|
|
|
|
| Use of residual |
|
|
|
|
| MIDI |
|
|
|
|
| Duan and Pardo |
|
|
|
|
| Li et al. |
|
|
|
|
melody_mix2, 6 dB 
|
|
source 1 |
source 2 |
source 3 |
residual |
| Original |
|
|
|
|
| Automatic |
|
|
|
|
| Disentangling |
|
|
|
|
| Use of residual |
|
|
|
|
| MIDI |
|
|
|
|
| Duan and Pardo |
|
|
|
|
| Li et al. |
|
|
|
|
melody_mix2, 1O dB 
|
|
source 1 |
source 2 |
source 3 |
residual |
| Original |
|
|
|
|
| Automatic |
|
|
|
|
| Disentangling |
|
|
|
|
| Use of residual |
|
|
|
|
| MIDI |
|
|
|
|
| Duan and Pardo |
|
|
|
|
| Li et al. |
|
|
|
|
7 violins mix 
|
|
source 1 |
source 2 |
source 3 |
source 4 |
source 5 |
source 6 |
source 7 |
residual |
| Original |
|
|
|
|
|
|
|
| No F0 track disentangling/No ideal grouping |
|
|
|
|
|
|
|
|
| No F0 track disentangling/Ideal grouping |
|
|
|
|
|
|
|
|
| F0 track disentangling/Ideal grouping |
|
|
|
|
|
|
|
|
| F0 track disentangling/No ideal grouping |
|
|
|
|
|
|
|
|
| MIDI-informed separation |
|
|
|
|
|
|
|
|
Thesis
G. Siamantas, "An iterative, residual-based approach to
unsupervised musical source separation in single-channel mixtures",
PhD thesis,
Dept. of Electronics, University of York, UK, 2009. [
PDF]
Contact
Giorgos Siamantas
g[DOT]siamantas[AT]gmail[DOT]com