Improve accuracy of Speech Recognition for different Indian Accents using MFCC, LPC, Zero Crossing and Power Spectrum
Keywords:MFCC, LPC, Zero Crossing, Power Spectrum. Support Vector Machine (SVM)
Nowadays, smart phones and its novel applications are widely used by different users. Among those
applications, speech recognition has been the most fascinating and useful application. In smart phones user gives
speech/spoken words as input, command given is interpreted and relevant task is carried out. The recognition of the
speech for different living in different countries is a challenging job. Observations show that most smart phones provide
more accurate and efficient results to American users rather than users from other part of the world. When these smart
phones are used in foreign countries like India, Japan etc. it gives less accurate results. This is due to the fact that the
speech recognition system used in smart phones uses majority of data from American origin in order to training it.
According to survey more and more people are using smart phones in India in recent times.The work undertaken uses an
accent base approach to improve the accuracy. Speech recognition system is developed for each accent and users can
acquire them based on their origin. For training purpose 3 accents are considered – Gujarati,Bengali and Malayalam.
Speech Recognition Features used in this are MFCC, LPC, Zero Crossing and Power Spectrum.SVM is used for