Product release ·
Voice Models on Singularity
A free, fully local alternative to cloud dictation apps like Wispr Flow, built into Singularity Desktop. Press a hotkey, speak, and your words land wherever you are typing. No cloud, no account, no subscription.
How it works
Press a key. Speak. Done.
Press Right Option and start talking. Press it again and the text appears at your cursor, in any app: your editor, your browser, a chat window, a terminal.
Your clipboard is put back the way it was, and password fields are never touched. You can change the hotkey, and Cmd+Shift+Esc cancels a take.
- 01
Pick a voice model
Choose one in Singularity Desktop. It downloads once and runs on your Mac from then on.
- 02
Hold the thought, not the mic
Right Option starts and stops. The default model shows your words while you are still speaking.
- 03
Text where you need it
Dictation is typed into whatever app has focus. Allow Accessibility once so it can type for you.
Models
Parakeet, Whisper, Apple Speech, and more.
Every model runs on your machine. Pick speed, accuracy or languages, and switch any time.
- Parakeet Realtime EOU 120M
- Default. English. Shows words live as you speak, every 160 ms. 224 MB.
- Parakeet TDT 0.6B v2
- English. Higher accuracy, with custom-word boosting for names and jargon. 567 MB.
- Whisper Tiny, Base, Small
- Nearly 100 languages, detected automatically. 78 MB, 148 MB and 488 MB.
- Apple Speech
- Built into macOS 14 and later. Nothing to download.
- Coming soon
- Parakeet TDT v3, Nemotron, Cohere Transcribe and Apple SpeechAnalyzer.
Parakeet models need a Mac with Apple silicon. Whisper models are multilingual; the Parakeet models are English only.
Private by default
Your voice never leaves your Mac.
Local dictation needs no wallet, no API key and no internet connection. Your history stays on your device.
It costs nothing to use, because nothing runs on our servers.
Singularity Grid
Serve transcription and earn.
The same app can put your Mac to work. Turn on serving and your machine transcribes audio for Singularity Grid users, and you keep 80% of what they pay.
Developers call it like any OpenAI-style speech endpoint, with an API key and credits.
- Model on the Grid
- whisper-1, through the OpenAI-style /v1/audio/transcriptions API
- Price to callers
- $0.006 per minute of audio ($0.0001 per second)
- Operator share
- 80% of every request your Mac serves
- Privacy
- Audio is end-to-end encrypted from the caller to the node that transcribes it
- Limits
- Up to 60 seconds of audio per request
- To serve
- A Mac with Apple silicon on macOS 15 or later, running Singularity Desktop
Availability
One app for everything local AI.
Voice Models ship today in Singularity Desktop for macOS. Windows support is on the way.
