Comparison of Techniques with Speech Enhancement and Nonlinear Rectification for Robust Speaker Identificatio

Silva H.A. · Proceedings - 2022 3rd International Conference on Electronics, Communications and Information Technology, CECIT 2022 (2022) · DOI: 10.1109/cecit58139.2022.00038

Resumen

Automatic speaker recognition is about the identification of a person based on his or her characteristic voice with the help of machines. However, this identification is drastically degraded due to adverse conditions. Our aim of this study is to evaluate the performance of the techniques Power- Normalized Cepstral Coefficients (PNCC) and Mel Frequency Cepstral Coefficients (MFCC) on NIST 2002 database, as well as using Wavelet Denoising and Cubic Law as techniques to speech enhancement and nonlinear rectification to improve speaker recognition rates. Results showed that combined Wavelet Denoising and Cubic Law get improved the recognition rates under noisy conditions.

Instituciones: Pontificia Universidad Católica del Perú