Skip to main content
Live transcription shows a running transcript in the overlay while you talk. The text is still inserted into your app once, after you stop recording, but the final insert is close to instant because most of the work is already done.

​
Which local models stream

Models built for streaming show text as you speak. The rest transcribe in one batch after you stop. The catalog marks this per model, and you can filter by it on the Models page. Models that stream include the Parakeet and Nemotron streaming variants and the Moonshine family. Whisper, Canary, Cohere, SenseVoice, GigaAM, and non-streaming Parakeet do not. See Transcription models for the full catalog. Switching to a non-streaming model moves you back to batch transcription automatically.

​
Cloud streaming

Two cloud providers support live streaming: ElevenLabs (with scribe_v2_realtime) and Deepgram. Their text arrives while you are still talking, so releasing the keys only flushes the last few words. Turn this on with the Transcribe as I speak switch in the cloud transcription settings. For ElevenLabs you need to select a realtime model; the batch model (scribe_v2) only transcribes after you stop. If the live stream fails or produces nothing, SpeakoFlow falls back to transcribing the full recording in one batch request. You never lose your words. The batch and realtime models can produce slightly different wording for the same audio, and the substitution is logged. See Cloud transcription for setup.

​
How it connects to the overlay

There is no separate streaming switch for local models. It follows the overlay style in Settings → Dictation → Dictation overlay.
  • Live. Streaming on. Shows a readable card with the running text.
  • Minimal. Streaming off. Shows the compact pill.
  • None. No overlay, no streaming.
A fresh install is on Auto, which resolves to Live when the selected model streams and Minimal when it does not. Picking any option writes a concrete choice that overrides Auto. If Live is selected but the model cannot stream, the overlay falls back to Minimal. A non-streaming model never shows an empty live card.

​
Two shades of text

The running text renders in two shades. Committed text is solid and will not change. A still-revisable tail is dimmer. Not every engine reports a revisable tail, so on some models everything you see is already committed.

​
The live card

The card is interactive: you can scroll back through the text, select a phrase, and copy it. It never takes keyboard focus from the app you are dictating into. See The recording overlay for more on the card’s behavior.
Last modified on August 30, 2026