Active Noise Cancellation Using a Wiener Filter

Author: heychandur
Language: MATLAB
Current version: V0.5

Offline audio-file noise reduction using an STFT-based Wiener-filter pipeline. The project evolves from basic audio inspection to configurable noise reduction with preview, refilter, and save controls.

Scope: This project processes a recorded audio file. It is not real-time active noise cancellation with a separate reference microphone.

Versions

Version	Purpose	Script
V0.1	Load, inspect, plot, and play audio	v01_load_and_play_audio.m
V0.2	Analyze waveform and average spectrum	v02_noise_analysis.m
V0.3	Initial frame-based Wiener-filter experiment	v03_wiener_noise_reduction.m
V0.4	Filtering with preview/save/refilter workflow	v04_preview_refilter_save.m
V0.5	Current version: fixed five-second noise profile	v05_wiener_noise_reduction.m

Features in V0.5

• Load supported audio files with MATLAB's audioread.
• Convert multichannel input to mono and remove DC offset.
• Apply a 55 Hz high-pass filter and optional 50 Hz mains-hum harmonic notch filters.
• Estimate a noise profile from the first five seconds of the current input.
• Apply STFT-based Wiener filtering, gain smoothing, spectral gating, level correction, and peak protection.
• Play the input and filtered output in sequence before asking what to do.
• Choose Save, Refilter, or Cancel after preview.
• Refilter the most recently processed output, not the original recording.
• Display waveform and spectrogram comparisons.
• Prevent saving over the selected source file.

Requirements

• MATLAB with Signal Processing Toolbox functions used by the scripts, including butter, filtfilt, stft, istft, hamming, and spectrogram.
• An audio recording supported by MATLAB's audioread.
• For V0.5's default settings, an input of at least five seconds.

Run V0.5

1. Clone or download this repository.
2. Open MATLAB and set the Current Folder to src/v0.5-configurable-noise-profile.
3. Run v05_wiener_noise_reduction.m.
4. Select the noisy audio file.
5. Listen to the original and processed previews.
6. Choose Save to export the result, Refilter to run another pass, or Cancel to exit without saving.

The save dialog supports WAV and MP3 choices; actual encoding support depends on the MATLAB version and system. WAV is a good default for compatibility.

Processing overview

1. Read and validate the selected recording.
2. Convert the input to mono and remove its mean.
3. Apply high-pass filtering and optional mains-hum notch filters.
4. Estimate noise from the first five seconds of the current pass.
5. Apply the STFT-based Wiener gain and spectral gate.
6. Reconstruct the output, apply limited level correction, and prevent excessive peaks.
7. Play the comparison, then save, refilter, or cancel.

Important limitations

• The first five seconds are used as the noise estimate and should contain mostly background noise. Speech or music in this segment may be suppressed.
• Repeated refiltering can damage desired audio. Listen to each pass before saving.
• Stereo input is reduced to mono, so stereo spatial information is not retained.
• Lower output volume alone does not prove better noise reduction. Evaluate both by listening and, where possible, against a clean reference.
• Performance has not been claimed as validated on every MATLAB release. Run the script in your target MATLAB environment.

Project structure

.
├── docs/
│   └── version-history.md
├── src/
│   ├── v0.1-basic-audio/
│   ├── v0.2-noise-analysis/
│   ├── v0.3-wiener-filter/
│   ├── v0.4-preview-refilter-save/
│   └── v0.5-configurable-noise-profile/
├── .gitignore
├── LICENSE
├── README.md
└── VERSION.md

See VERSION.md for the current version and docs/version-history.md for the milestone-by-milestone history.

Author

heychandur — GitHub profile