Acoustic fingerprinting in research
Joren Six - IPEM, Ghent University - joren.six@ugent.bePresentation on: 0110.be/a/berlin
- Applications
- Implementation
- Reproducibility
- Conclusions
Applications of acoustic fingerprinting
Compare meta-data
Twins
| First twin | Second twin | |
|---|---|---|
| Audio | ||
| Year recorded | ? | 1949 |
| Title | The daughter Mandega | ? |
| People | Zezuru | Shona / Zezuru |
| Collector | Hugh Tracey | Hugh Tracey |
Improve listening experiences
Re-use segmentation
Synchronize sensors
Synchronize sensors
Synchronize sensors
Implementation: Panako
Needs
- Granular
- 'Scalable'
- Robust to speed changes
No systems were available (or described in academic literature).
Panako
- Acoustic fingerprinting system
- Apply in research
- Research platform to experiment with fingerprinting algorithms
Panako
Is inspired on- Published Sazam algorithm
- Granular
- Scalable
- A symbolic music fingerprinter by Andreas Artz
- Robust against time-stretching
- Sebastien Fenet's fingerprinter
- Robust against pitch-shifting
Panako
- Exact hashing
- Aligned matches
- False positives minimized
- Reports pitch-shift and time-stretch percentages
Panako practical
- Audio decoded via ffmpeg
- Constant-Q based - using TarsosDSP
- MapDB storage
- Portable Java
Computational research and reproducibility
Papers should be reproducibile.
Papers should be reproduced.
Problems
- Code remains unpublished
- Copyrighted material as test-data
- Lack of incentives
Replicating a paper
- Checks described method (code or text)
- Checks evaluation (data and procedure)
- Checks results
Replication results
System behaves similar as original but differences are unexplained
- Implementation bugs?
- Evaluation (data or procedure)
Conclusion
- Acoustic finterprinting techniques have many applications
- Panako tries to offer a practical system
- Reproducible methods catalyze research