AUTO-ADMIN: Non-MP3, WAV, OGG, SoundCloud, YouTube, Vimeo, Twitter and Facebook links in this post have been protected automatically. Once the member reaches 5 posts the links will function as normal.
Hi all,I’ve just released an open-source Android app that applies real-time adaptive equalization to all system audio — including Spotify, YouTube, games, podcasts, etc. — using a hybrid approach of on-device machine learning and native C++ DSP.
Key features:Works as a global audio effect (no per-app integration needed)
Uses a 25 KB neural model (Tiny AutoFUS) to predict per-band EQ gains
Applies them via a biquad filter cascade (8 bands, adjustable Q/freq/gain)
Runs 100% offline, no root, no internet
Latency: ~30 ms on mid-range devices (Snapdragon 7+)
Includes WAV export for calibration sweeps
Tech stack:C++/NDK for low-latency DSP (BiquadFilter, FFT, NoiseGate)
Kotlin for Android integration (AudioEffect, service lifecycle)
PyTorch Mobile for model inference
MIT licensed, full source available
GitHub: https://github.com/Kretski/audio-optimizer-android (https://github.com/Kretski/audio-optimizer-android)
(APK demo included in Releases)
Questions for the community:Any tips for smoothing biquad coefficient updates to avoid artifacts?
How do OEMs like Samsung or Sony handle global audio effects — any known quirks?
Would a log-spaced band layout be more perceptually effective?
This is part of a broader edge-AI initiative (AZURO), but the audio module is fully standalone and designed for reuse.
Thanks for any feedback!