Improving Phoneme Discrimination with Phonetic Motor Invariants
Author Information
Author(s): Claudio Castellini, Leonardo Badino, Giorgio Metta, Giulio Sandini, Michele Tavella, Mirko Grimaldi, Luciano Fadiga
Primary Institution: LIRA-Lab, University of Genova, Italy
Hypothesis
Can the use of phonetic motor invariants improve automatic phoneme discrimination?
Conclusion
Phonetic motor invariants significantly enhance the ability to automatically discriminate between bilabial and dental consonants, especially in noisy conditions.
Supporting Evidence
- Motor invariants were found to be more effective than traditional audio features in phoneme discrimination tasks.
- The study demonstrated that reconstructed motor features maintained performance even as noise levels increased.
- Results indicated that motor features provide a more invariant representation across different speakers and coarticulating phonemes.
Takeaway
The study shows that using information about how we physically make sounds can help computers understand speech better, especially when there's background noise.
Methodology
The study used a multi-subject database of synchronized speech and motor trajectories to identify phonetic motor invariants and tested their effectiveness in a neural network-based classifier.
Potential Biases
Potential bias due to the limited demographic of participants, all being female Italian speakers.
Limitations
The study's findings may not generalize to all phonemes or languages, and the sample size was limited to six speakers.
Participant Demographics
Six female Italian native speakers.
Statistical Information
P-Value
p<0.05
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website