Improved Speech Recognition using Vector Quantization, Genetic Algorithm and Neural Networks

Speech recognition has long been one of the most difficult and active research problems in Artificial Intelligence (AI). Automatic Speech Recognition (ASR) finds applications in the field of military, transportation, healthcare, electronic systems and many more. This technology enables humans to communicate with a computer interface in a manner that resembles normal human conversation. This paper proposes an improved automatic speech recognition using Vector Quantization, Genetic Algorithms and Neural Networks. A genetic algorithm is used for feature selection from an ordered set of speech data generated using vector quantization. The neural network further is used for the recognition of speech data. For validation of the improved method data is collected in which a line is spoken in five languages by 25 speakers. Experimental results demonstrate that the improved method achieves 93% accuracy with the minimum number of features.