Real-time Transcription
Process supported transcription models while recording and show optional live text.
Enable Real-time in a Mode's Transcription section when its model supports streaming. Partial transcripts may change as you continue speaking; the final text still goes through formatting, replacements, and optional enhancement after recording ends.
Supported Models
| Models | Real-time behavior |
|---|---|
| Parakeet V2, V3, Ultra, Unified | Optional |
| Nemotron Latin, Nemotron Multilingual | Required for live recording |
| Deepgram Nova 3 and Nova 3 Medical | Optional |
| ElevenLabs Scribe V2 | Optional |
| Mistral Voxtral | Optional |
| Gemini 3.5 Transcribe | Optional |
| Soniox V5 | Optional |
| Speechmatics | Optional |
| AssemblyAI Universal-3.5 Pro | Optional |
| xAI Grok Voice Transcribe 2.0 | Optional |
| Cartesia Ink 2 | Required; streaming-only and English-only |
Local Whisper, Apple Speech, Cohere Transcribe, SenseVoice, Groq Whisper, AssemblyAI Universal-2, OpenRouter transcription, and custom transcription endpoints do not provide this streaming option.
Show or Hide Partial Text
Settings → Interface → Live Text Display controls whether partial text is visible in the recorder. Hiding the display does not turn off real-time processing in the Mode.
Both Notch and Mini recorder styles can display live text. Cloud streaming sends audio to the selected provider during recording and needs a working connection.
Imported Files
The Transcribe view uses batch processing rather than live streaming. A Mode's Real-time setting does not make streaming-only providers support files; Cartesia Ink 2 cannot process imported files.