Voice Preprocessing and Feature Extraction for Speech Samples

This repository contains a Python implementation of a voice preprocessing workflow and feature extraction for machine learning (ML) applications. The project uses audio samples from the LibriSpeech test-clean dataset and demonstrates key steps in preparing speech data for ML/DL models. The workflow includes audio loading, preprocessing, visualization, feature extraction (MFCC), and a bonus experiment with noise addition.

Features

Automatic Dataset Download Downloads the LibriSpeech test-clean subset (around 330 MB). Extracts .flac audio files for processing. Audio Preprocessing Pipeline Resampling to 16 kHz Normalization Silence trimming Optional low-pass filtering

Visualization

Plots raw and processed waveforms side-by-side. Visualizes MFCC feature heatmaps for both audio samples.

Feature Extraction

Extracts Mel-Frequency Cepstral Coefficients (MFCCs). Handles variable audio lengths by padding MFCCs for tensor compatibility. Combines features into a single tensor ready for ML/DL models. Noise Experiment (Bonus) Adds Gaussian noise to audio samples. Demonstrates how preprocessing improves noisy audio.

Requirements

Python 3.9+ Libraries: librosa, matplotlib, scipy, torch, numpy

You can install all dependencies via pip:

pip install librosa matplotlib scipy torch numpy