This project implements a speech codec using Linear Predictive Coding (LPC) with a focus on audio compression and synthesis. It was developed as part of a graduate-level audio signal processing course at Johns Hopkins University.

The project performs:

• LPC analysis and encoding of speech signals
• Pitch and voicing detection using Modified Autocorrelation (MACF)
• Synthesis using LPC filter, glottal pulse models, and voiced/unvoiced detection

• lpc_main.m: Main pipeline for loading audio, encoding, decoding, and output.
• lpc_encoder.m: LPC coefficient extraction, pitch detection, voicing classification.
• lpc_decoder.m: Speech reconstruction using glottal pulse models or white noise.
• lpc_rosenberg.m: Generates glottal pulses based on the Rosenberg model.
• results/: Includes plots comparing original and synthesized signals.
• audio_samples/: All .wav files used for input/output testing.

Adjustable via global variables in lpc_main.m:

• CLIP_RATIO = 0.60
• ORDER_VOICED = 20
• VOICE_THRESH = 0.35
• MAX_PITCH = 450 (adjusted for female pitch range)

Plots of the original vs synthesized audio demonstrate intelligibility and the LPC filter envelope tracking the signal formants.

Requires MATLAB. Audio toolbox recommended.