Author : J.Rexy 1
Date of Publication :29th March 2018
Abstract: The social network is the main ambassador for information sharing in this digital era. Nowadays social network is being utilized by different level of people in diverse platforms for many reasons.Most of the data which are shared by social media are unstructured data which does not have a predefined format. There are many examples of unstructured data; among those audio plays, a vital role as many audios are being shared in Social Media. Audio data can be utilized to leverage intelligence such as identify the speaker, identifying emotions, identifying the area of talk etc. Since Tamil is one of the longest surviving classical languages in the world and very few researchers are focused on Tamil audio analysis, this paper deals with Tamil audio speaker recognition implemented in Matlab version 13. The basics of speech recognition, feature extraction process, and pattern matching pave the way for identifying and classifying Tamil audios according to speakers are also reviewed. In order to improve the efficiency of the recognition process, during the pre-processing stage, Adaptive Weiner filter is employed for removing unwanted noise from the audio signal. After the pre-processing stage, the retrieved enhanced signal is utilized for feature extraction process which is carried out using combined LPC (Linear Predictive Analysis) and Mel- Frequency Cepstral Coefficients (MFCC). The mfcc coefficients are used as audio classification features to improve the classification accuracy. LPC is one of the most powerful speech analysis techniques and is a useful method for encoding quality speech at a low bit rate. Hence MFCC and LPC could contribute more to extract best features.In order to increase the accuracy rate of training and recognition,MFCC and LPC are combined in feature extraction.The feature extraction process generates feature vectors which are extracted for further processing. The extracted feature vectors are applied to hybrid MLP and SVM machine learning Algorithm to identify the speaker and classify the audios accordingly
Reference :
-
- Diment, Aleksandr, and Tuomas Virtanen. "Archetypal analysis for audio dictionary learning." In Applications of Signal Processing to Audio and Acoustics (WASPAA), 2015 IEEE Workshop on, pp. 1-5. IEEE, 2015
- Gaston,Daniel,Martin and Marta.”Feature Analysis for Audio Classificaion” In: Bayro-Corrochano E., Hancock E. (eds) Progress in Pattern Recognition, Image Analysis, Computer Vision, and Applications. CIARP 2014. Lecture Notes in Computer Science, vol 8827. Springer, Cham
- Abd El-Fattah, M. A., Moawad Ibrahim Dessouky, Salah M. Diab, and Fathi El-SayedAbd El-Samie. "Speech enhancement using an adaptive wiener filtering approach." Progress In Electromagnetics Research M 4 (2008): 167- 184.
- Leena R Mehta 1, S.P.Mahajan 2, Amol S Dabhade,”Comapritive study of MFCCand LPC for Marathi isolated word recognition system” International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering, Vol. 2, Issue 6, June 2013.