Emotional Speech Recognition using Gammatone Cepstral Coefficients

Ekta garg , Madhu Bahl

ABSTRACT The key challenge in speech emotion recognition system is determining emotion from noisy speech, because extraction of emotion gets complicated because of background noise. This complication leads to the mismatching between Training and Testing calculation. Thus, quality of feature extraction algorithm plays an important role. Therefore, to enhance the robustness of the system, gammatone cepstral coefficients (GTCC) is being proposed. This technique is based on gammatone filter bank which attempts to mimic the human auditory system. When GTCC is compared to Mel Frequency Cepstral Coefficients (MFCC), the time-domain GTCC provides better recognition performance under all noise as well as clean conditions. MFCC is a conventional feature extraction technique which does not perform well under noisy conditions. GTCC captures speakerâ€™s characteristics and discards irrelevant characteristics. A comparison is done between GTCC and MFCC on the basis of Signal to noise (SNR) and Mean Square Error (MSE) Parameters. Further their performance is evaluated using Feed Forward Back Propagation Neural Network which is a supervised machine learning method. Neural networks are used to make quick and accurate recognizing of emotion. Classification is done using Training phase and testing phase. From the results it is concluded that proposed GTCC algorithm enhance the performance with higher SNR and lower MSE. Keywords:- Feature extraction, Gammatone Cepstral Coefficients, Emotion Recognition, Back Propagation Neural Network

Title:	Emotional Speech Recognition using Gammatone Cepstral Coefficients
Author Name:	Ekta garg , Madhu Bahl
Abstract:	ABSTRACT The key challenge in speech emotion recognition system is determining emotion from noisy speech, because extraction of emotion gets complicated because of background noise. This complication leads to the mismatching between Training and Testing calculation. Thus, quality of feature extraction algorithm plays an important role. Therefore, to enhance the robustness of the system, gammatone cepstral coefficients (GTCC) is being proposed. This technique is based on gammatone filter bank which attempts to mimic the human auditory system. When GTCC is compared to Mel Frequency Cepstral Coefficients (MFCC), the time-domain GTCC provides better recognition performance under all noise as well as clean conditions. MFCC is a conventional feature extraction technique which does not perform well under noisy conditions. GTCC captures speakerâ€™s characteristics and discards irrelevant characteristics. A comparison is done between GTCC and MFCC on the basis of Signal to noise (SNR) and Mean Square Error (MSE) Parameters. Further their performance is evaluated using Feed Forward Back Propagation Neural Network which is a supervised machine learning method. Neural networks are used to make quick and accurate recognizing of emotion. Classification is done using Training phase and testing phase. From the results it is concluded that proposed GTCC algorithm enhance the performance with higher SNR and lower MSE. Keywords:- Feature extraction, Gammatone Cepstral Coefficients, Emotion Recognition, Back Propagation Neural Network
Cite this article:	Ekta garg , Madhu Bahl , " Emotional Speech Recognition using Gammatone Cepstral Coefficients " , International Journal of Application or Innovation in Engineering & Management (IJAIEM), Volume 3, Issue 10, October 2014 , pp. 285-291 , ISSN 2319 - 4847.
Full Text [PDF] Home