We’ve all been there: you record a great take, a podcast, or a tutorial, only to realize the hum of your PC fan or the hiss of your audio interface has "colored" the entire track. I tried the standard one-click solutions, but they often left my voice sounding metallic or "underwater."
So, I decided to build my own solution: AuraDSP.
🛠 The Problem: Spectral Artifacting
Most consumer-grade denoisers use simple gates. If the signal is below a threshold, they cut it. The problem? Background noise lives inside the same frequencies as your voice. If you cut the frequency, you kill the "body" of the audio.
The Solution: AuraDSP’s Architecture
I focused on a Spectral Subtraction model. Instead of just cutting volume, the app analyzes the noise floor and mathematically subtracts that specific profile from the active signal.
Key Engineering Features:
- STFT Analysis Window: Uses a 2048-sample analysis window to balance frequency resolution with real-time latency.
- The "Noise Delta" Monitor: I added a toggle that lets you hear only what is being removed. This is critical for ensuring you aren't accidentally subtracting the "warmth" of a vocal.
- 5th-Order Butterworth Filtration: To handle sub-sonic rumble (trucks passing outside, desk bumps), I integrated a steep 80Hz high-pass filter.
- Dual-Process Core: To prevent UI lag during heavy DSP processing, the app uses a dedicated Engine core that communicates with a lightweight Launcher.
📊 Professional Visualization
I wanted to see the noise to understand it. AuraDSP features a high-resolution Spectrogram (TFT). It maps the "Aura" of the audio in real-time, allowing you to visually identify 50/60Hz electrical hums or high-frequency digital whine that the human ear might miss during long sessions.
🏥 Beyond Content Creation
While I built this for YouTubers and podcasters, the precision of the subtraction logic makes it interesting for more sensitive applications, such as medical audio analysis or forensic recovery, where hearing the "delta" (the difference) is just as important as the final output.
Try the Standalone Build
I am releasing the binaries (EXEs) for testing. I’m looking for feedback on the "artifact-to-suppression" ratio across different microphones and environments.
- Zero Dependencies: No runtimes or installers required.
- High Transparency: Designed to keep transients sharp while killing the floor.
Check out the project and the technical documentation here:
👉 [https://github.com/shafqatgreat/AuraDSP]
I’d love to hear your thoughts on the suppression logic—especially from those of you working in untreated rooms!
Top comments (0)