Beamforming based Speech Recognition using Genetic Algorithm for Real-time Systems
Milind U. Nemade1, Satish K. Shah2
1Milind U. Nemade, Department of Electronics and Telecommunication, University of Mumbai, K.J. Somaiya Institute of Engg. Information Technology, Mumbai (Maharashtra), India.
2Prof. Satish K. Shah, Department of Electrical Engineering, M.S. University of Baroda, Faculty of Technology and Engineering, Vadodara (Gujrat), India.
Manuscript received on 21 May 2013 | Revised Manuscript received on 28 May 2013 | Manuscript published on 30 May 2013 | PP: 96-104 | Volume-2 Issue-2, May 2013 | Retrieval Number: B0618052213/2013©BEIESP
Open Access | Ethics and Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: The speech based applications have been always important in communication for the humans. There are in various essential applications like speech recognition, voice-distance-talk and other forms of personal communications. Most recently, speech based interface has been tried to be employed in almost all the mobile and stationary devices. However, these attempts could not give ultimate response due to variations in surrounding noises, changes in person to person speech and also intra person variation. This scenario leads to further research that will make speech recognition more robust and general and can be applied upcoming electronic devices to be sued for gaming, entertainment, cellular phones. The broad categories of speech enhancement techniques can be listed as speech filtering techniques, beam forming techniques and active noise cancellation methods. In this paper, we have improved the performance of beamforming based speech recognition system using evolutionary computational algorithms (Genetic algorithm, GA). Additionally, the system is made to be working in real-time as time required for classifier has been reduced dramatically. This is particularly achieved by including the zeros at random places and in random amount in initial population chromosomes, which were generated randomly in the range of 0 to 1. This results in the reduction of feature elements in feature descriptor and have feature vector length. The experiments were performed for 20 words including numbers and commands, 10 words of numbers only and 10 words of commands only for different values of filter bank parameters. The results show the effectiveness of the GA optimization in all the subsets of experiments with different parameters of beamforming.
Keywords: Delay and Sum Beamformer, HMM Based Classifier, Least Mean Square, MFC, Nearest Neighbor Classifier.
Scope of the Article: Classification