Mark your audio
Embed the mark locally on your hardware.
Prove the audio is AI-generated.
Declare it as yours.
Beta phase until December 2nd, 2026!
Authenticity
Open tools can embed a mark, but no one stands behind it. A Cobalstamp mark is verified by a detector that recognizes it, us.
Your audio is trusted.
A mark that can't be forged
Each mark carries a cryptographic integrity code tied to the provider that issued it. No one else can issue a mark in your name.
No one else can mark as you.
The mark names its provider
Every mark can be linked to a provider ID. Detection can tell you exactly who produced it.
You know the source of any audio in the wild.
EU AI Act, Article 50
Non-compliance can be fined up to €15 million or 3% of global turnover. Cobalstamp is the route to compliance by covering the marking and the detection of watermarks.
The route to compliance.
Self-serve, no sales call
No enterprise contract, no sales cycle. Sign up, mark, detect — same day.
You're marking today, not next quarter.
Tie your own data to the mark
Each mark carries a unique ID. Associate it with a project, a client, a license — in your own database.
Every mark becomes a key to your records.
SDK for audio watermarking
Technical features
- Marking, on your hardware
- Very low memory usage
- No dynamic memory allocation
- No GPU needed
- On average 1000x faster than real-time on consumer hardware
- Known and constant latency before processing: 40 ms (compensated for when using file-based processing)
- Detection, on our servers
- API
- Free
EU AI Act Article 50
- Article 50(2) (providers): you mark your audio with Cobalstamp SDK. Cobalstamp API provides the detection mechanism.
- Article 50(4) (deployers): Cobalstamp's detection tells you which content is AI-generated or manipulated, so you know what to disclose.
Audio features
- Can process any audio signals/recordings (speech, music, etc.)
- Imperceptible distortions
Quick setup
Follow these steps to register and run a quick example.
Install the Python package
Create an account
Create an API key
Set the API key in your environment
Start using the Python package
You're all set!
Run a quick example
Load audio, add a Cobalstamp watermark, and check the watermark back with a detection call. Save the following in a .py file and run it with python:
Pricing
License model: You pay a monthly fee, you mark as much as you want on your side. We handle the detection requests on our side. Overage beyond the plan quota is free during the beta phase.
Testing Marker
Test the full pipeline on your own audio, no credit card required, no commitment.
- Limited to 1h of marking
- Python SDK
- File-based marking and detection
- No parallel execution
- No commercial deployment
- Marks forgotten after one month
- No support for EU AI Act Art. 50 compliance
Marker Startup
For products that are live and still growing.
- 100h of marking per month
- Overage 0.001€/min
- Route to EU AI Act compliance
- Python SDK
- File-based marking and detection
- No parallel execution
Marker Studio
For operations in production with time-critical workflows.
- All of Marker Startup plus:
- 1000h of marking per month
- Unlimited parallel marking and detection
- Available mark ID for data association
- Priority detection queue
Detector
- Python SDK
- File-based detection
- Unlimited detection
- No marking
Custom
- All of Marker Studio plus:
- C, C++, Swift SDK
- Android SDK
- iOS SDK
- Streaming encoding
- Real-time encoding
- Integration services
- Custom support
| Marker Startup | Marker Studio | Custom | |
|---|---|---|---|
| Marking | |||
| Included marking | 100h/mo | 1000h/mo | On request |
| Overage | 0.001€/min | 0.001€/min | On request |
| Parallel execution | |||
| Detection | |||
| Detection | |||
| Mark ID for data association | |||
| Priority detection queue | |||
| SDK | |||
| File-based marking | |||
| Python SDK | |||
| Linux/MacOS/Windows SDK | |||
| C, C++, Swift SDK | |||
| Android SDK | |||
| iOS SDK | |||
| Support | |||
| Integration services | |||
| Custom support | |||