AhmadHassan-BTed/NaSuno

AI audio watermark removal & forensic anti-attribution toolkit. Neutralizes Suno, Udio, and neural fingerprints with ISO 226 perceptual masking.

What it solves

NaSuno addresses the issue of "forensic tagging" in AI-generated audio. Platforms like Suno, Udio, and ElevenLabs embed invisible watermarks, ultrasonic carriers, and metadata into their generated audio files to track provenance. This project provides tools to remove these markers, allowing creators to use AI-generated stems or textures in their own compositions without being automatically flagged or attributed to those platforms.

How it works

NaSuno uses a combination of digital signal processing (DSP) and psychoacoustic modeling to surgically remove tracking markers while preserving audio quality. It employs techniques such as zero-phase IIR notch filtering to eliminate ultrasonic carriers (like 19.5 kHz tones), ISO 226 equal-loudness contours to ensure modifications remain inaudible, and bitstream structural excision to strip hidden metadata (like C2PA or ID3 tags). The processing is designed to be transparent, maintaining the original transient punch and stereo integrity of the audio.

Who it’s for

This project is for musicians, sound designers, and audio producers who incorporate AI-generated elements into their work and wish to maintain creative privacy and avoid automated platform surveillance or copyright attribution.

Highlights

  • Comprehensive Removal: Targets 14 distinct forensic vectors, including ultrasonic carriers, phase-locked micro-artifacts, and hidden metadata chunks.
  • Psychoacoustic Integrity: Uses Bark-scale critical band analysis to ensure anti-attribution measures are mathematically inaudible.
  • Web-Based Studio: Includes a client-side Web Audio DSP studio that allows users to process files directly in the browser without uploading data.
  • Transparent Processing: Employs zero-phase filtering to prevent phase smearing or ringing, ensuring the original audio quality is preserved.

Written about in

Related

  • Project
    Picovoice/porcupine

    A lightweight and highly accurate wake word engine for building always-listening voice-enabled applications across IoT, mobile, and web platforms.

  • Project
    microsoft/muzic

    A research project from Microsoft Research Asia that provides a suite of deep learning models for music understanding, generation, and AI-driven music processing.

  • Project
    wenet-e2e/wenet

    WeNet is a production-ready end-to-end speech recognition toolkit that provides full-stack solutions, achieving state-of-the-art results on public datasets while being easy to install and use.

  • Project
    remsky/Kokoro-FastAPI

    Dockerized FastAPI server that wraps the open‑source Kokoro‑82M TTS model, offering an OpenAI‑compatible speech endpoint, streaming, multilingual voices, voice‑mixing, inline SSML tags, custom voice‑tuning and a web UI. Runs on CPU, NVIDIA/AMD GPUs or Apple Silicon via Docker or uv.