Speech recognition using arithmetic coding and MFCC for Telugu language

This paper presents speech feature extraction of Telugu language through proper compression. Compression is provided to speech using Digital Arithmetic coding and features are extracted by MFCC then classification is done by ANN. Speech feature extraction and feature classification are the major steps in ASR. This paper presents a technique to extract the speech features after speech compression. A technique with arithmetic coding and MFCC is done by reducing the average number of bits. Arithmetic coding and MFCC stands out in terms of magnificence and potency. A text dependent Telugu ASR is designed. Features extraction process is done for 140 bits/frame and 80 bits /frame and features extracted are LSP, Pitch prediction filter, code base indexes, gain, synchronization, FEC, future expansion. The proposed technique AC with MFCC has been compared with various existing techniques like ADPCM, LD-CELP, CS-AELP, CELP, LPC, and MFCC. The performance of the proposed technique has proved better in terms of bit rate, word error rate and compression ratio.