|
| EPSRC Reference: |
GR/S85900/01 |
| Title: |
Advanced Subband Systems for Audio Source Separation |
| Principal Investigator: |
Professor M Plumbley |
| Other Investigators: |
|
| Researcher Co-investigator: |
|
| Project Partner: |
|
| Department: |
Sch of Electronic Eng & Computer Science |
| Organisation: |
Queen Mary, University of London |
| Scheme: |
Standard Research |
| Starts: |
01 June 2004 |
Ends: |
29 February 2008 |
Value (£): |
176,431
|
| EPSRC Research Topic Classifications: |
| Digital Signal Processing |
|
|
| EPSRC Industrial Sector Classifications: |
| Aerospace, Defence and Marine |
Communications |
|
| Related Grants: |
|
| Panel History: |
|
|
Summary |
Humans primarily use phase Information to localize sounds at low frequency, whereas In the upper frequencies Intensity differences dominate due to inherent phase ambiguities. We aim to exploit this Idea to develop new adaptive BSS algorithms that are robust to source/sensor movement and noise. These algorithms will sacrifice exact phase reconstruction at high frequency for algorithm robustness. However rather than trying to simply mimic human audition we will take an engineering approach and Identify the sensitivity of phase and amplitude within the problem and thus determine the best solutions within our computational framework. We will apply these new methods to both blind and semi-blind (ones that exploit known sensor Information) adaptive systems and all algorithms will be thoroughly evaluated, both objectively and subjectively to determine performance and robustness
The project will also focus on the computational cost of the BSS algorithms. For such systems to be useful In future electronic devices, such as digital hearing aids or tele-conferencing systems It Is Important that the solutions can be Implemented In a real time manner. To this end we will build a real time prototype system to test our new Ideas.
|
| Final Report Summary |
|
Humans primarily use phase information to localize sounds at low frequency, whereas in the upper frequencies intensity differences dominate due to inherent phase ambiguities. We aim to exploit this idea to develop new adaptive blind source separation (BSS) algorithms that are robust to source/sensor position and noise. These algorithms will sacrifice exact phase reconstruction at high frequency for algorithm robustness. However rather than trying to simply mimic human audition we will take an engineering approach and investigate the sensitivity of phase and amplitude within the problem and thus determine the best solutions within our computational framework. We will apply these new methods to both blind and semi-blind (ones that exploit known source and sensor information) adaptive systems and all algorithms will be thoroughly evaluated, both objectively and subjectively to determine performance and robustness. The project will also focus on the computational cost of the BSS algorithms. For such systems to be useful in future electronic devices, such as digital hearing aids or tele-conferencing systems it is important that the solutions can be implemented efficiently. To this end we will build a prototype system to test our new ideas.
|
| Further Information: |
|
| Organisation Website: |
http://www.qmul.ac.uk |
|
|