RealtimeSTT itself is released under the MIT license. Optional transcription, VAD, and wake-word engines pull in their own packages, model weights, service terms, and model cards — and those upstream terms can change independently from this repository. This page is a practical summary, not legal advice. Before shipping a product, redistributing model files, bundling wheels, or offering hosted transcription, review the upstream license files and terms for the exact package, model, revision, API, and distribution path you use.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/KoljaB/RealtimeSTT/llms.txt
Use this file to discover all available pages before exploring further.
Important Limits
- “Commercial use” in the tables below means the cited upstream license or model card does not obviously prohibit commercial use. It does not cover privacy, consent, biometrics, export controls, trademarks, data-processing agreements, or platform-specific terms.
- Runtime code and model weights can have different licenses.
- Hosted APIs and model hubs can add access terms even when the model weights are open-licensed.
- If you redistribute binaries, wheels, model files, or Docker images, preserve upstream copyright notices and license texts.
- For Hugging Face models, verify the license on the exact model repository and revision you download. Derived, converted, quantized, or ONNX/GGUF versions may carry the upstream model license plus additional notices.
ASR Engine Licenses
| RealtimeSTT engine | Runtime / package license | Default or typical model license | Commercial-use note |
|---|---|---|---|
faster_whisper | faster-whisper is MIT. Its inference runtime, CTranslate2, is MIT. | OpenAI Whisper code and model assets are published in the openai/whisper repository under MIT. | Generally permissive, subject to preserving MIT notices when redistributing. |
whisper_cpp | whisper.cpp is MIT. RealtimeSTT uses the pywhispercpp package; check that package metadata before redistributing wheels. | Uses OpenAI Whisper-family model files; check whether you use original OpenAI files or third-party conversions. | Generally permissive for original Whisper assets, but converted model files should be checked individually. |
openai_whisper | openai-whisper is MIT. | OpenAI Whisper assets in the same repository are MIT. | Generally permissive, subject to MIT notice preservation. |
sherpa_onnx_moonshine, sherpa_onnx_parakeet | sherpa-onnx is Apache-2.0. ONNX Runtime is MIT. | The model bundle determines the model license. Useful Sensors Moonshine Streaming is MIT; NVIDIA Parakeet is CC-BY-4.0. | Keep the license and attribution files that ship with the selected sherpa-onnx model bundle. Do not assume every bundle has the same terms. |
kroko_onnx | kroko-onnx is Apache-2.0. | Kroko Community models are CC-BY-SA; Commercial/OEM models are separately licensed. | CC-BY-SA-style licenses generally require attribution and share-alike. Verify commercial fit with Kroko before using Community models commercially; free license keys are described as non-commercial. |
parakeet | NVIDIA NeMo source is Apache-2.0; some NeMo Framework containers and deployment artifacts have separate NVIDIA product terms. | The default nvidia/parakeet-tdt-0.6b-v3 model card lists CC-BY-4.0 and is ready for commercial/non-commercial use. | CC-BY-4.0 requires attribution. Review NVIDIA container/product terms if deploying through NIM, Riva, NeMo containers, or other NVIDIA services. |
omnilingual_asr | Meta’s omnilingual-asr package and model suite are documented as Apache-2.0. | The Meta Omnilingual ASR code and models are Apache-2.0; Meta documents the accompanying corpus as CC-BY-4.0. | The model license is permissive. If you use or redistribute corpus data, handle the corpus license separately. |
moonshine, moonshine_streaming | Uses Hugging Face Transformers in RealtimeSTT; check the installed Transformers version and any model custom code. | The default UsefulSensors/moonshine-streaming-medium model card lists MIT. | English Moonshine Streaming weights are permissive. Non-default Moonshine variants may differ; check the exact model card. |
granite_speech, granite | Uses Hugging Face Transformers in RealtimeSTT. | The default ibm-granite/granite-speech-4.1-2b model card lists Apache-2.0; IBM’s Granite docs describe Granite Speech as Apache-2.0. | Generally permissive. Preserve Apache-2.0 notices when redistributing. |
qwen3_asr, qwen_asr | Uses the qwen-asr package and, optionally, vLLM. Check those package licenses for redistribution. | The default Qwen/Qwen3-ASR-1.7B model card lists Apache-2.0. | Generally permissive for the default model. vLLM, Docker, ModelScope, or hosted serving paths may add their own terms. |
cohere_transcribe, cohere | Uses Hugging Face Transformers and model-specific code from the model repository. | The default CohereLabs/cohere-transcribe-03-2026 model card lists Apache-2.0 and describes the model as an open-source release. | Generally permissive for the model card license. If access is gated or served through a provider, follow that provider’s access terms too. |
openai_api | Placeholder engine in this repository; not wired as a local ASR backend. | N/A | If implemented later, usage would be governed by OpenAI API/service terms, not an open-source model license. |
VAD and Wake-Word Licenses
| Component | Upstream terms | Practical note |
|---|---|---|
WebRTC VAD via webrtcvad-wheels | PyPI lists MIT. | Permissive wrapper; preserve notices if redistributing. |
| Silero VAD | silero-vad is MIT. | Permissive for the packaged VAD assets, subject to MIT notices. |
| OpenWakeWord | OpenWakeWord is documented as Apache-2.0. | Permissive wake-word path. Verify any third-party wake-word model you add. |
| Porcupine / Picovoice | Picovoice publishes separate service and SDK terms. | Treat Porcupine as a provider-licensed dependency. Free trials or free plans may not cover all commercial deployment scenarios. |
Key Callouts
Whisper Family
Faster-whisper, whisper.cpp, and openai-whisper all use OpenAI Whisper model weights published under MIT. Generally permissive for commercial use, subject to preserving MIT copyright notices.
OpenWakeWord
OpenWakeWord is Apache 2.0. Permissive for commercial use. Verify any third-party wake-word model file you add beyond the defaults.
Kroko / Banafo Models
Community models are CC-BY-SA (attribution and share-alike required). Pro and Commercial/OEM models are separately licensed and require a license key. Do not store license keys in code or committed files.
Porcupine
Porcupine is a commercial product with a free tier. The free tier may not cover production or commercial deployments. Review Picovoice’s terms before using it in a product.
Source Links
The following upstream sources were checked when compiling this page (last researched: 2026-05-21):- Faster Whisper: https://github.com/SYSTRAN/faster-whisper/blob/master/LICENSE
- CTranslate2: https://github.com/OpenNMT/CTranslate2
- OpenAI Whisper: https://github.com/openai/whisper/blob/main/LICENSE
- whisper.cpp: https://github.com/ggml-org/whisper.cpp
- sherpa-onnx: https://github.com/k2-fsa/sherpa-onnx
- ONNX Runtime: https://github.com/microsoft/onnxruntime
- kroko-onnx: https://github.com/kroko-ai/kroko-onnx
- Kroko on-premise model/license docs: https://docs.kroko.ai/on-premise/
- NVIDIA NeMo: https://github.com/NVIDIA/NeMo
- NVIDIA NeMo Framework license notes: https://docs.nvidia.com/nemo-framework/user-guide/25.11/overview.html
- NVIDIA Parakeet model card: https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
- Meta Omnilingual ASR: https://github.com/facebookresearch/omnilingual-asr
- Meta Omnilingual ASR blog: https://ai.meta.com/blog/omnilingual-asr-advancing-automatic-speech-recognition/
- Useful Sensors Moonshine Streaming Medium model card: https://huggingface.co/UsefulSensors/moonshine-streaming-medium
- IBM Granite Speech model card: https://huggingface.co/ibm-granite/granite-speech-4.1-2b
- IBM Granite Speech docs: https://www.ibm.com/granite/docs/models/speech
- Qwen3-ASR model card: https://huggingface.co/Qwen/Qwen3-ASR-1.7B
- Cohere Transcribe model card: https://huggingface.co/CohereLabs/cohere-transcribe-03-2026
- webrtcvad-wheels PyPI metadata: https://pypi.org/project/webrtcvad-wheels/
- Silero VAD license: https://github.com/snakers4/silero-vad/blob/master/LICENSE
- OpenWakeWord project page: https://openwakeword.com/
- Picovoice terms: https://picovoice.ai/docs/terms-of-use/