Related Experiment Videos
Fractal dimensions of speech sounds: computation and application to automatic speech recognition
1Department of Electrical and Computer Engineering, National Technical University of Athens, Greece. maragos@cs.ntua.gr
The Journal of the Acoustical Society of America
|March 25, 1999
Summary
This study quantifies speech turbulence using fractal models, developing an algorithm to analyze speech signals. This fractal dimension analysis offers modest improvements in automatic speech recognition systems.
Area of Science:
- Acoustics and Signal Processing
- Computational Linguistics
- Fractal Geometry
Background:
- Speech production involves airflow dynamics that can create turbulence.
- Quantifying this turbulence geometry is challenging.
- Fractal models offer a potential method for analysis.
Purpose of the Study:
- To quantify speech turbulence geometry using fractal models.
- To develop an efficient algorithm for estimating the short-time fractal dimension of speech signals.
- To explore the application of fractal dimension in speech segmentation, phonetic classification, and automatic speech recognition.
Main Methods:
- Utilized fractal models to quantify speech turbulence.
- Developed an efficient algorithm for estimating short-time fractal dimension based on multiscale morphological filtering.
- Integrated short-time fractal dimension at multiple scales as features in hidden Markov model-based automatic speech recognition systems.
Main Results:
- An efficient algorithm for estimating short-time fractal dimension was described.
- The potential of fractal dimension for speech segmentation and phonetic classification was discussed.
- A modest improvement in speech recognition performance was achieved using fractal dimension features.
Conclusions:
- Fractal models provide a viable method for quantifying speech turbulence.
- Multiscale morphological filtering enables efficient fractal dimension estimation.
- Fractal dimension features can enhance automatic speech recognition systems.