EXPLORE PUBLICATIONS BY COUNTRIES

Voice banking, Product Code Vector Quantizers, Linear Predictive Coefficients, Line Spectral Frequencies

Voice banking is an excellent telephone banking service by which a user can access his account for any service at any time of a day, in a year. The speech techniques involved in voice banking are speech coding and speech recognition. This paper investigates the performance of a speech recognizer for a coded output at 20 bits/frame obtained by using various vector quantization techniques namely Split Vector Quantization, Multi Stage Vector Quantization, Split-Multi Stage Vector Quantization, Switched Split Vector Quantization using Hard decision scheme, Switched Multi Stage Vector Quantization using Soft decision scheme and Multi Switched Split Vector Quantization using Hard decision scheme techniques. The speech recognition technique used for recognition of the coded speech signal is the Hidden Markov Model technique and the speech enhancement technique used for enhancing the coded speech signal is the Spectral Subtraction technique. The performance of vector quantization is measured in terms of spectral distortion in decibels, computational complexity in Kflops/frame, and memory requirements in floats. The performance of the speech recognizer for coded outputs at 20 bits/frame has been examined and it is found that the speech recognizer has better percentage probability of recognition for the coded output obtained using Multi Switched Split Vector Quantization using Hard decision scheme. It is also found that the probability of recognition for various coding techniques has been varied from 80% to 100%.