Paper
Selected Topics on Time-Frequency Matrix Decomposition Analysis
- Authors:
- Behnaz Ghoraani
- Abstract
- While time-frequency (TF) representations are commonly used for visualization purposes, the adaptive feature extraction from TF matrices has not been extensively studied in the literature. Even when used, there exists no unique or automated methodology to extract discriminative TF features from non-stationary signals. The present paper focuses on feature extraction from time-frequency distribution (TFD), and attempts to develop a generalized TF matrix (TFM) analysis methodology that exploits the benefits of TFD in pattern classification systems. The proposed TFM feature extraction consists of two stages: first, we build TFM decomposition that uses a matrix decomposition (MD) technique to effectively segment non-stationary signals. Second, instantaneous and unique features are extracted from each segment in a way that they successfully represent joint TF structure of the signal. In this paper, the suitable tools for TFM feature extraction pertaining to pattern classification are investigated through experiments with different synthetic signals. Experiments performed with pathological speech and environmental audio signals produced 98.6% and 85.5% accuracy rates respectively demonstrating the benefits of TFM feature extraction for pattern classification related applications.
- Keywords
- Time-Frequency Matrix Analysis; Time-Frequency Analysis; Time-Frequency Feature Extraction; Matrix Decomposition
- StartPage
- 64
- EndPage
- 78
- Doi