Applying Dictionary Learning Algorithms In Sparse Representation of Speech Data

Articles in Press

Document Type : Research Paper

Authors

1 Lorestan University

2 Department of Electrical engineering,, lorestan university. Khorramabad

3 Electrical Engineering, Faculty of Engineering and Technology, University of Mazandaran, Babolsar, Iran

Abstract

As a widely used technique in signal processing, Sparse representation has gained significant attention in various fields, including data compression, noise reduction in speech and image signals, pattern recognition, and other signal processing-related issues. In such representations, signals are linearly combined using a small number of dictionary atoms, leading to data dimensionality reduction and improved signal processing efficiency. To accurately represent speech data, an appropriate dictionary is required to effectively represent speech signals' characteristics. In this paper, dictionaries are trained using dictionary learning algorithms and sparse representations such as MOD, K-SVD, RAMC, UD4-MOD, and OMP, in the time, time-frequency, and wavelet transform domains. The performance of the obtained dictionaries is evaluated using various time-frequency metrics such as RE, MSE, fwSegSNR, SegSNR, PESQ, and STOI. The results demonstrate that employing the K-SVD dictionary learning algorithm in conjunction with the OMP sparse representation algorithm in the STFT domain achieves promising results for speech signal reconstruction.

Keywords

Main Subjects

Journal of Modeling in Engineering

Articles in Press, Accepted Manuscript
Available Online from 14 September 2025

Files

History

  • Receive Date: 07 May 2024
  • Revise Date: 29 March 2025
  • Accept Date: 24 June 2025

Share

How to cite

Statistics