Internship | Noise subtraction for robust acoustic event classification
AQCURA
AQCURA is an automated acoustic quality control solution designed to identify manufacturing defects and assess product quality at the end of a production line. By combining advanced signal-processing and machine-learning techniques with Microflown particle velocity sensors, products can be tested directly in noisy manufacturing environments.
A particle velocity probe positioned in the near field of the device under test provides strong natural rejection of environmental noise, particularly at low and mid frequencies. At higher frequencies, however, background noise can still interfere with the measured signal and reduce the reliability of acoustic event classification.
Project description
The objective of this project is to develop and validate noise-subtraction techniques that improve acoustic event classification under low signal-to-noise conditions.
The proposed measurement configuration combines a PU probe positioned close to the device under test with an additional reference microphone located farther away from the device. The PU probe primarily captures the acoustic radiation of the device, while the reference microphone provides information about the surrounding background-noise field.
The student will investigate signal-processing methods capable of separating or suppressing independent acoustic components contained in the two measurement signals. Possible approaches include adaptive noise cancellation, Wiener filtering, spectral subtraction, coherence-based filtering, transfer-function estimation, blind source separation and time-frequency masking. Other suitable methods may also be explored.
If the work and findings are of good quality, preparing a publication for an international conference or scientific journal will be encouraged. In summary, the work should include a review of relevant noise-reduction methods, algorithm development, computer simulations, experimental testing and a quantitative assessment of the classification results.
Requirements
The candidate should:
- be proficient in programming with MATLAB and/or Python;
- have a good understanding of digital signal processing;
- have prior knowledge of audio processing, acoustic measurements or noise-control techniques;
- be comfortable carrying out experimental measurements and analysing data;
- be able to work independently and document the developed methods and results clearly.
Experience with adaptive filtering, source separation, time-frequency analysis or real-time signal-processing implementation would be an advantage.
Particle Velocity Sensors
AQCURA | Acoustic Quality Control Testing
End of Line testing software