Speech Noise Cancellation Based on a Neuro-Fuzzy System: Further Improvements
Anna Esposito*,**, Eugene C. Ezin*,*** and Carlos A. Reyes-Garcia*,****
*International Institute for Advanced Scientific Studies (IIASS) Via G. Pellegrino 19, 84019, Vietri sul Mare, SA, Italy
**INFM Salerno University
***Institut de Mathematiques et de Sciences Physiques (IMSP) B.P. 613 Porto-Novo, Benin,
****Instituto Tecnologico de Apizaco-COSNET Apizaco, Tlaxcala, Mexico 90300
This work reports on an experimental system based upon the Adaptive Neuro-Fuzzy Inference System (ANFIS) architecture, which is employed for identifying a nonlinear model of the unknown dynamic characteristics of the noise transmission paths. The output of this model is used to subtract the noisy components from the received signal. The novelty of the system described in the present paper, with respect to our previous work, consists in a different set up, which requires more fuzzy rules, generated by seven trapezoidal membership functions, and uses a second order it sinc function to generate the nonlinear distortion of the noise. Once trained for few epochs (only three) with a long sentence corrupted with babble noise, the FIS obtained, has the ability to clean speech sentences corrupted by babble and also by car, traffic, and white noise, in a computational time almost close to realtime. The average improvement, in terms of SNR, was 37 dB without further training.
This article is published under a Creative Commons Attribution-NoDerivatives 4.0 International License.