Design of a Real-Time Speech Recognition System using CNN for Consumer Electronics

Modern systems such as consumer electronics, automotive electronics, domestic appliances, music systems, air conditioners, televisions are becoming smarter with built-in voice controlled features, enabling hands-free operation. Real-time speech recognition is heart of such systems and research is in progress towards enhancing these systems to be functional in Indian languages too. Design and evaluation of a real-time speech recognition system using Convolution neural networks for Kannada language is proposed here. Performance of proposed system is evaluated using samples recorded in the lab, as standard speech datasets for this language are not available. A maximum recognition accuracy of 99.60% is obtained on using the proposed system and details pertaining to steps followed to enhance recognition accuracy of proposed system are also reported. The proposed system can be easily extended to other Indian and foreign languages.