Design of a linear predictive speech coder

A speech coder was designed on MATLAB environment based on human speech production system that uses linear prediction as mathematical tool. System does not work in real time. It processes the recorded speech signals by coding with linear prediction (analysis), writes to a file in a predefined format with .lpc extension. Then it reads and decodes the file (synthesis), plays the input-output signal and shows some visual animations. System can work 3.84 kb/s or 2.56 kb/s data output rates. A subset of the Turtel Library (5 male and 5 female speaker, 15 sentences of each) provided by National Electronic and Cryptology Research Institute and 3.84 kb/s data rated vocoder was used in tests. Sound quality of the vocoder is defined as 3.4 with Absolute Category Rating test by participation of 8 men and 7 women audiences whose average age is 31. Keywords: Model based speech coder, linear prediction, LPC, MATLAB