README
Welcome to Kroko 👋
Open-source speech recognition built for developers.
Our engine is fully open-source, and you choose how to deploy models: use our CC-BY-SA licensed community models or upgrade to commercial models with premium performance. We focus on building fast, high-quality production models and providing examples that take the guesswork out of integration.
Why Kroko ASR?
- ⚡ Fast & lightweight – optimized Zipformer models (Whisper and parakeet style coming).
- 🧩 Flexible licensing – use fully open-source CC-BY-SA community models or integrate commercial/OEM models for premium accuracy.
- 🌍 Runs anywhere – cross-platform and with support for many programming languages.
- 📱 Mobile & web ready – works on Android, (iOS coming soon) in the browser via WASM, and with WebSockets for streaming.
- 🧰 Production focus – we prioritize real-world performance, stability, and examples.
- 🤝 Customizable – bring your own model, fine-tune for domain-specific vocabularies, or commission us.
Our mission: fast, high-quality ASR with licensing that works for both open-source and closed-source projects.
Demos
▶️ Android App
Run speech recognition natively on your phone using ONNX Runtime.
🌐 Browser (WASM)
Experience transcription directly in your browser, no server required.
Models
Kroko ASR follows a unique dual-model strategy:
1. Community Models (free, open-source)
- Licensed under CC-BY-SA.
- Low-latency, lightweight models.
- Perfect for hobby projects, research, or free tiers.
- Faster and smaller than Whisper/Parakeet in many scenarios.
2. Commercial & OEM Models
- Premium accuracy and robustness.
- Licensed for professional and production products.
- Designed for SaaS, dev tools, and enterprise integration.
3. Bring, Train, or Commission Your Own
- DIY: Use our training guides to build and distribute your own models.
- Professional services: Work with us to create fine-tuned models for accents, jargon, or specialized domains.
This gives you full freedom: start free, scale commercially, or roll your own.
Our Community
Join the Kroko community to learn, share, and contribute:
- 💬 Discord – chat with developers, ask questions, and share projects.
- 📢 Reddit – join discussions, showcase your integrations, and follow updates.
- 🤗 Hugging Face – explore our models, try live demos, and contribute feedback.
Contributing
PRs welcome! Run ruff, black, and pytest before submitting.
License
Apache-2.0 engine. Models licensed separately (CC-BY-SA community or commercial OEM).
Credits
Kroko ASR is built on top of Sherpa-ONNX.
⚠️ Note: Kroko ASR is an independent project and is not affiliated with Sherpa-ONNX. We build on their excellent open-source engine, but our models, demos, and packaging are developed and maintained separately.
Contributors
README
Welcome to Kroko 👋
Open-source speech recognition built for developers.
Our engine is fully open-source, and you choose how to deploy models: use our CC-BY-SA licensed community models or upgrade to commercial models with premium performance. We focus on building fast, high-quality production models and providing examples that take the guesswork out of integration.
Why Kroko ASR?
- ⚡ Fast & lightweight – optimized Zipformer models (Whisper and parakeet style coming).
- 🧩 Flexible licensing – use fully open-source CC-BY-SA community models or integrate commercial/OEM models for premium accuracy.
- 🌍 Runs anywhere – cross-platform and with support for many programming languages.
- 📱 Mobile & web ready – works on Android, (iOS coming soon) in the browser via WASM, and with WebSockets for streaming.
- 🧰 Production focus – we prioritize real-world performance, stability, and examples.
- 🤝 Customizable – bring your own model, fine-tune for domain-specific vocabularies, or commission us.
Our mission: fast, high-quality ASR with licensing that works for both open-source and closed-source projects.
Demos
▶️ Android App
Run speech recognition natively on your phone using ONNX Runtime.
🌐 Browser (WASM)
Experience transcription directly in your browser, no server required.
Models
Kroko ASR follows a unique dual-model strategy:
1. Community Models (free, open-source)
- Licensed under CC-BY-SA.
- Low-latency, lightweight models.
- Perfect for hobby projects, research, or free tiers.
- Faster and smaller than Whisper/Parakeet in many scenarios.
2. Commercial & OEM Models
- Premium accuracy and robustness.
- Licensed for professional and production products.
- Designed for SaaS, dev tools, and enterprise integration.
3. Bring, Train, or Commission Your Own
- DIY: Use our training guides to build and distribute your own models.
- Professional services: Work with us to create fine-tuned models for accents, jargon, or specialized domains.
This gives you full freedom: start free, scale commercially, or roll your own.
Our Community
Join the Kroko community to learn, share, and contribute:
- 💬 Discord – chat with developers, ask questions, and share projects.
- 📢 Reddit – join discussions, showcase your integrations, and follow updates.
- 🤗 Hugging Face – explore our models, try live demos, and contribute feedback.
Contributing
PRs welcome! Run ruff, black, and pytest before submitting.
License
Apache-2.0 engine. Models licensed separately (CC-BY-SA community or commercial OEM).
Credits
Kroko ASR is built on top of Sherpa-ONNX.
⚠️ Note: Kroko ASR is an independent project and is not affiliated with Sherpa-ONNX. We build on their excellent open-source engine, but our models, demos, and packaging are developed and maintained separately.