Seed funding for strategic research @ RTP (research manpower­)

• Main Author: Thambipillai Srikanthan.
• Other Authors: School of Computer Engineering
• Format: Research Report
• Language: English
• Published: 2009
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Pattern recognition
• Online Access: http://hdl.handle.net/10356/17242
• Institution: Nanyang Technological University

Modern voice authentication systems perform extremely well on large population high quality clean speech databases. While novel algorithms can be designed to provide performance and accuracy, the performance degrades rapidly in the presence of noise. Noise introduces a mismatch between the verification utterance and the speaker template which causes unpredictable scores leading to performance degradation. Our research attempts to address the problem of mismatch condition caused by additive noise. We have proposed novel algorithms for noise compensation in the speaker model domain and demonstrated their efficiency on TIMIT database corrupted with additive noise. Subsequently, we have combined the proposed algorithm with spectral subtraction method to further improve the performance of the authentication have been successfully translated to dedicated hardware architecture and prototyped on FPGA-based platforms.