audio signal
- 1. User-Driven Quality Enhancement for Audio Signal Processing Convention Paper 8823 D. Comminiello, S. Scardapane, M. Scarpiniti, A. Uncini 134th AES CONVENTION Rome, Italy – 2013 May 4-7
- 2. Outline of the Talk User-Driven Quality Enhancement 2 2013 May 4-7 (1) • Problem Definition • Our Framework (2) • Applications • Immersive Speech, Games… (3) • Preliminary Results • Conclusions
- 3. Understanding the User User-Driven Quality Enhancement 3 2013 May 4-7 Audio Processor Personal Judgement Development Stage ???
- 4. Evaluation Procedures User-Driven Quality Enhancement 4 2013 May 4-7 Objective Indexes Psychoacustic Models Subjective Tests
- 5. Classical Approach User-Driven Quality Enhancement 5 2013 May 4-7
- 6. Our Approach User-Driven Quality Enhancement 6 2013 May 4-7 Possible Enhanced Audio
- 7. Interactive Evolutionary Algorithm User-Driven Quality Enhancement 7 2013 May 4-7 Pool of Possible Settings Subjective Evaluation «Reproduction» Selection New Pool
- 8. Main Drawbacks User-Driven Quality Enhancement 8 2013 May 4-7 User Fatigue User Discrimination Fast convergence obtained with few possible fitness values Time Constraint Partial Ordering
- 9. Our Proposal User-Driven Quality Enhancement 9 2013 May 4-7 1. IEC has been seldomly used in audio processing applications. 2. We believe it to be of practical interest for a wide range of tasks. 3. This is the main reason we are proposing this framework.
- 10. Applications – Games User-Driven Quality Enhancement 10 2013 May 4-7 Audio for Games
- 11. Applications – Forensic Audio User-Driven Quality Enhancement 11 2013 May 4-7 Forensic Audio (Image property of SoundAndSound)
- 12. Applications – Immersive Experience User-Driven Quality Enhancement 12 2013 May 4-7 Immersive Audio (Image property of Integrated Media Systems Center)
- 13. Interactive AEC User-Driven Quality Enhancement 13 2013 May 4-7
- 14. Test Setup User-Driven Quality Enhancement 14 2013 May 4-7 1. Five signals distorted by female voice. 2. Echo cancellation through affine projection algorithm (APA) with 4 parameters. 3. A standard GA minimizes normalized misalignment: 7. An IGA should minimize user’s preferences.
- 15. Test Setup - Workflow User-Driven Quality Enhancement 15 2013 May 4-7 Original Signals IEC Minimization Set 1 Standard Minimization Set 2 Subjective Comparison
- 16. Results User-Driven Quality Enhancement 16 2013 May 4-7 55% 28% 17% Preferences IGA GA UNKNOWN
- 17. Conclusions User-Driven Quality Enhancement 17 2013 May 4-7 1. Good results of our framework on AEC. 2. User fatigue is the main drawback to be confronted. 3. Several applications awaits in the future. 4. Possible combination of objective and subjective measurements.