Volume 54, Number 6, November-December 2020
|Page(s)||1555 - 1572|
|Published online||16 September 2020|
Multiple objective optimization Applied to Speech enhancement problem
USTHB University, Lab. AMCD&RO, DGRSDT, Faculty of Mathematics, Department of Operations Research Bab-Ezzouar, BP32 El-Alia, 16122 Algiers, Algeria
2 École Supérieure Ali CHABATI Reghaia, Algiers, Algeria
Accepted: 10 November 2019
Enhancement of speech corrupted by broadband noise is subject of interest in many applications. For several years, the investigation of methods of denoising the vocal signal has yielded very satisfactory results, but certain problems and questions still remain. The term speech quality in speech enhancement is associated with clarity and intelligibility. So, one of these issues is to reach a compromise between noise reduction, signal distortion and musical noise. In this paper, we studied one of the classical techniques based on the spectral subtraction developed by Boll and improved by Berouti where two parameters α and β to control the effects of the distortion and the musical noise are introduced. However, the choice on these parameters (α and β) remains empirical. Our works is to find a compromise between these two parameters to obtain an optimal solution depending on the environment, the unknown noise and its level. Moreover, we propose in this paper, an algorithm based on bi-objective approach precisely Particle Swarm Optimization (PSO) technique in association with speech enhancement technique proposed by Berouti et al. Comparative results show that the performance of our proposed method with several types of noise, depending on the environment and on various noise levels, are better than those of spectral subtraction methods of Boll or Berouti.
Mathematics Subject Classification: 90C05 / 90C25 / 90C29 / 65K05
Key words: Speech enhancement / spectral subtraction / multiobjective optimization / meta-heuristic / PSO
© EDP Sciences, ROADEF, SMAI 2020
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.