Cadenza
Open-source voice input for macOS
Hold a key, speak, and the text appears at your cursor. Recognition runs locally, not on someone else's server.
NewNew in 1.2.0: voice translation, more cloud services, quiet speechFeatures
Small, quiet, and yours.
Voice typing
Hold a key, speak, let go: the text lands at your cursor. Hesitations and stutters are tidied, and your vocabulary fixes names and jargon.
Voice translationNew
Hold a second shortcut and the translation is typed instead. Use any AI service you add, including Ollama on this Mac.
Cloud, if you want itNew
Your own account with OpenAI, Groq, Google, Azure, AssemblyAI, ElevenLabs or Chinese services. Audio is sent only after you agree.
Quiet speechNew
Local models clean up soft recordings in a noisy room, so a whisper is not lost.
Screenshots and text recognition
Capture, mark up, pin, and copy the text or QR code in a picture. On this Mac by default.
For developers and AI hardware
An optional local API turns speech from your own programs, pendants or glasses into text.
Privacy
Local first. Cloud only if you ask.
Local
- SenseVoice
- FireRedASR2
- Parakeet
Nothing is uploaded.
Your own cloud account
- OpenAI
- Groq
- Google Cloud
- Azure
- AssemblyAI
- ElevenLabs
- iFLYTEK
- Volcengine
- Tencent Cloud
- Alibaba Cloud
- Baidu
- Deepgram
Audio is sent only after you agree, provider by provider.
No account, no analytics, no crash reports. A "never go online" switch hides every option that could connect. Read the privacy promise
Open source
Read it. Build it. Make it better.
Cadenza is MIT licensed and developed in the open. What leaves your Mac is there in the code for anyone to check.
$git clone https://github.com/DragonKingIO/Cadenza-voice.git$cd Cadenza-voice$./cadenza/tools/fetch-sherpa-onnx.sh$./cadenza/build.sh --stage-only