Adjusted Levenberg - Marquardt technique for improvement speech recognition system

This paper proposes the adjusted Levenberg-Marquardt(LM) technique to improve the performance of speech recognition. Number of difference neuron in hidden layer has been trialed in verities input speech signal which generated from 40 speakers: 20 males and 20 females. The results show that the system provides difference values of the accuracy percentage depending on the number of neuron in hidden layer. The highest accuracy rate for trained and untrained speech data of female approved at 96.67% and 98.33% for stress consonant respectively and also provided 86.92% and 90.00% for voiced consonant of male.