Linear Predictive Coding for Speech Compression

Telecommunication industry is growing and different services are rapidly introduced by different competitors to attract the users. Speech communication and its quality conservation is the most prevalent and common service provided by almost all companies. The objective of this project is the development of a LPC (Linear Predictive Coding) based voice coder. Attributes for speech like pitch, voiced and unvoiced decision and silence were extracted and speech was modeled using LDR (Levinson Durbin Recursion) and SDA (Steepest Descent Algorithm). LPC filter is analyzed and its model is implemented. LPC's different attributes complexity, delay and bitrate are deliberated and tradeoffs are highlighted. The results were analyzed and quality of speech was determined using spectrograph and by listening to the synthesized speech. At the end quality of original and synthesized speech is discussed and shown graphically and a soft comparison between both above mentioned technique is also added