An educational repository showcasing digital signal processing (DSP) and machine learning techniques to isolate, analyze, and remove unwanted noise from audio signals. This project was developed as part of the Engineers Australia UOWD ECTE Workshop.
This project explores three core methodologies to clean noisy audio signals, transitioning from standard mathematical smoothing filters to adaptive AI-powered restoration models.
- Moving Average Filter: Smooths out random high-frequency signal spikes by calculating localized averages across a defined window size.
- Low-Pass Filter (LPF): Attenuates sharp, static background noise by blocking frequencies above a set threshold while preserving foundational music frequencies.
- AI-Based Linear Regression Model: Trains an adaptive linear regression model using feature matrices (including multiple moving averages) to map relationship patterns between clean and noisy audio data, predicting a restored audio output.
ai_audio_cleaning.m— Predictive AI model using Linear Regression (fitlm) and localized multi-window signal smoothing.low_pass_filter.m— Implementation of frequency-selective filtration using traditional low-pass filtering.moving_average_filter.m— Implementation of a moving window average smoothing technique.non_noise.wav— Clean baseline audio file used for training the AI model.noisy_music.wav— The original corrupted audio file containing background noise.