This repository contains the implementation and report for a Linear Predictive Coding (LPC)-based Speech Synthesizer, developed for the Speech & Audio Processing module during the MSc in Artificial Intelligence at the University of Surrey.
To synthesize vowel sounds (“ea” and “oo”) by simulating the Source–Filter model of speech production:
- Source: Impulse train simulating vocal cord excitation
- Filter: Linear Predictive model simulating vocal tract response
| File | Description |
|---|---|
EEEM030_Assignment1.ipynb |
Complete Python implementation |
EEEM030_Assignment1_Report.docx |
Full academic report with results & visuals |
assets/ |
📸 Screenshots from waveform and spectrum plots |
| LPC Order | Quality | Notes |
|---|---|---|
| 6 | Robotic | Limited formant detail |
| 10 | Clearer | Better vowel shape, less noise |
| 20 | ✅ Best | Most natural result; preserved formants |
| Formant | Male (Hz) | Female (Hz) |
|---|---|---|
| F1 | 328.13 | 679.69 |
| F2 | 1875.00 | 2320.31 |
| F3 | 3492.19 | 3562.50 |
Male Segment:
Female Segment:
pip install librosa numpy matplotlib scipy