Abstract:
The performance of speaker verification systems in mobile devices may degrade significantly when speech corrupted by noise environment. To improve the performance of spea...Show MoreMetadata
Abstract:
The performance of speaker verification systems in mobile devices may degrade significantly when speech corrupted by noise environment. To improve the performance of speaker verification system under noise mobile environment. we propose a new method of feature extraction algorithm. The new features called Multitaper Gammaton Hilbert Envelope Coefficients (MGHECs) which are defined based on a recently developed Mean Hilbert Envelope Coefficients (MHECs). The performance of speaker verification system using i-vector-GPLDA modeling. The performance of the proposed feature extraction was evaluated using National Institute of Standards and Technology (NIST) speaker recognition evaluation (SRE) 2008 data. Noise mobile environment simulated by passing the NIST 2008 database through Adaptative Multi-Rate (AMR) (WB : WideBand) speech codec and NOISEX92 database with different level SNR. The evaluation results demonstrate that the proposed feature extraction outperforms the conventional method (MHEC).
Date of Conference: 25-26 April 2018
Date Added to IEEE Xplore: 07 June 2018
ISBN Information: