FluidAudio is a Swift framework for on-device speaker diarization and audio processing, optimized for Apple’s Neural Engine to deliver performance more efficient than CPU or GPU-based models.
Built to support real-time workloads on iOS and older macOS devices, it fills the gap left by closed-source or CPU-only solutions, offering an open-source alternative that makes powerful voice AI features like speaker diarization and identification freely available.
Key Features
- State-of-the-art diarization
- Voice activity detection (VAD)
- CoreML models
- Real-time processing
- Cross-platform
Pros & Cons
| Pros | Cons |
| Research-competitive speaker separation with optimal speaker mapping. | Requires macOS 13.0+ and iOS 16.0+ for full support. |
| Generate speaker embeddings for voice comparison and clustering. You can use this for speaker identification. | It is limited to the CoreML ecosystem. |
| Designed for real-time workloads, but also works for offline processing. |
Pricing & Plans
The FluidAudio repository is licensed under Apache‑2.0, which is a permissive open-source license that allows free use, modification, and distribution, even in commercial projects.
You don’t have to pay to use the tool, but feel free to support the developers and contributors.
Conclusion
FluidAudio is a unique open-source, Swift-native framework for real-time, on-device speaker diarization, fully optimized for Apple’s Neural Engine.
Unlike tools like pyannote-audio, NeMo, or cloud APIs, it delivers efficient, private, offline processing for iOS and macOS, which fills the gaps that alternative tools still have.
