Speech coding using Linear Predictive Coding (LPC10)

Various speech compression techniques have been developed in the past. Human Speech frequency varies from 300 Hz to 3400 Hz speech compression, more commonly referred to as speech coding. It is a method by which the amount of information required to represent a speech signal can be reduced depending on the application requirement (Sridharan, Leis and Paliwal 1999). Usually Lossy algorithms are used for speech coding because of the fact that loss of speech quality is often detectable by the human ear. High correlation between adjacent speech samples can be considered advantageous as most speech compression is achieved using linear digital filters. One of the methods of compression used for the speech production is LPC (Linear Predictive Coding).The current thesis addresses LPC speech analysis and synthesis algorithm using Matlab environment. The script is written and simulated for a given input speech signal to reconstruct the output speech using Levinson Durbin coefficients defined for LPC10. The research focus used the .wav file as an input speech signal. Different waveforms have been observed on every phase of the encoding and decoding block. When connected to a microphone, we can also provide the input speech and the lossy algorithm as LPC reproduces the input speech on the other end of the decoder. The script can further be implemented using a DSP processer, such as TMS320CXX, and the script shall work fine using the processor as well. The matlab script for encoding and decoding of LPC10 coefficients have been tested for everyday sentences such as: "The Discrete Fourier transform of a real valued signal is conjugate symmetric" and other similar sentences. The code was implemented using the DSP processor kit and the reconstructed signal was recorded in the .wav format using the matlab function wavread.