Related Experiment Videos
Spectral-shape features versus formants as acoustic correlates for vowels
1Department of Electrical and Computer Engineering, Old Dominion University, Norfolk, Virginia 23529.
The Journal of the Acoustical Society of America
|October 1, 1993
Summary
Spectral shape features offer superior vowel classification compared to traditional formants, even when fundamental frequency is considered. Static features are most crucial for vowel discrimination, with dynamic information providing secondary benefits.
Area of Science:
- Acoustic Phonetics
- Speech Processing
- Machine Learning for Speech
Background:
- Formants, the first three spectral prominences, have historically been key acoustic cues for vowels.
- Spectral shape features offer a more comprehensive description of the vowel spectrum.
Purpose of the Study:
- To compare the effectiveness of formants versus spectral shape features for automatic vowel classification.
- To investigate the roles of static and time-varying acoustic information in vowel discrimination.
Main Methods:
- Automatic vowel classification experiments were conducted on monopthongal vowels in CVC words.
- Spectral shape was represented using cosine expansion coefficients of the scaled magnitude spectrum.
- Comparisons were made with and without the inclusion of fundamental frequency (F0) information.
Main Results:
- Spectral shape features outperformed formants in vowel classification across most conditions, especially without F0.
- Including F0 improved classification for both feature types but maintained the superiority of spectral shape.
- Perceptual confusion patterns more closely matched errors from spectral shape classification.
Conclusions:
- Spectral shape features provide a more complete set of acoustic correlates for vowel identity than formants.
- Static acoustic features are primary for vowel discrimination, while time-varying features offer valuable secondary information.