Kyutai STT: streaming speech-to-text models
Don’t want to install Kyutai STT? Drop the file here.
The first 20 minutes of any recording are transcribed free: a recording up to 20 minutes transcribes in full, a longer one is previewed, and a pack from $6 (500 minutes) transcribes the rest and opens every export, and credits never expire.
Two streaming models: stt-1b-en_fr (about 1B parameters, 0.5 s delay, semantic voice-activity detection) and stt-2.6b-en (English, 2.5 s delay), with word timestamps, under CC-BY-4.0.
Get Kyutai STT Official site & pricing
Kyutai STT streams transcripts as people speak: English and French (1B, 0.5 s delay) or English (2.6B, 2.5 s delay), word timestamps, CC-BY-4.0. Rust server for production, MLX for Apple devices.
From the team behind Moshi. For recordings that already exist, a file transcriber is the simpler tool.
What it is
Kyutai STT is a pair of decoder-only streaming speech-to-text models built on the delayed-streams-modeling approach behind Kyutai's Moshi. kyutai/stt-1b-en_fr has about 1B parameters, a 0.5-second delay and a semantic voice-activity detector useful for voice agents; kyutai/stt-2.6b-en has about 2.6B parameters and a 2.5-second delay, English only. They emit text as audio arrives, with capitalisation, punctuation and word timestamps, and the card reports good results on audio as long as 2 hours. Kyutai provides PyTorch for research, a Rust server for production (64 simultaneous connections at 3× real time on an L40S, 400 real-time streams on an H100) and MLX for Mac and iPhone. Weights are CC-BY-4.0; the repository code is Apache-2.0.
Watch out for: English and French only (2.6B is English-only). Tuned for real-time streams rather than file batches, and serving it means running the Rust server on a GPU.
For recordings, not live streams
- Drop the audio or video file into the box at the top of this page, or paste a link (YouTube, a podcast or RSS feed, Google Drive, Twitch or Kick). Nothing to install and no GPU needed.
- The language is detected automatically and the transcript stays in the language that was spoken (WhipScribe does not translate), including languages beyond English and French.
- Read the summary free. A pack from $6 (500 minutes) opens the full transcript and the downloads: TXT, DOCX, SRT, VTT and JSON. Packs are one-time purchases and credits never expire.
- Files can be up to 10 hours or 5 GB per file.
Related: Bulk transcription · SRT generator · Transcribe large files · WhisperX · Pricing
Install / use
pip install moshi # python -m moshi.run_inference --hf-repo kyutai/stt-2.6b-en audio.mp3Key specs
What people actually do with Kyutai STT-style transcription
The tool is the means. These are the jobs — each one priced at published rates, each one wired up on its own page.
Features
| Speaker diarization | Not stated |
| Word-level timestamps | Yes |
| Streaming / real-time | Yes |
| Languages supported | 2 |
| HIPAA eligible | No |
Links
- github.com/kyutai-labs/delayed-streams-modeling ↗PyTorch, Rust and MLX implementations · Apache-2.0 code
- kyutai/stt-1b-en_fr ↗~1B · English + French · CC-BY-4.0
- kyutai.org/next/stt ↗project page: semantic VAD, benchmarks
Kyutai STT vs Whipscribe
| Feature | Kyutai STT | Whipscribe |
|---|---|---|
| Category | Open source | Transcription APIs |
| Pricing | free (CC-BY-4.0 weights) | $6–$24 one-time packs (credits never expire) · first 20 minutes free |
| Speaker identification | No | Named in the summary, key quotes and chapters |
| Word timestamps | Yes | Yes |
| Streaming | Yes | No |
| Languages | 2 | 99 |
| Platforms | Linux, macOS, iOS | Web, API, MCP |
Alternatives to Kyutai STT
Frequently asked about Kyutai STT
Which Kyutai STT model should I use?
stt-1b-en_fr for English or French with low latency (0.5 s) and semantic voice-activity detection, which suits voice agents; stt-2.6b-en for English only, with a 2.5 s delay.
Is Kyutai STT the same as Moshi?
No. Moshi is Kyutai's speech-to-speech dialogue model; Kyutai STT is a speech-to-text model built on the same multistream approach.
How do I run Kyutai STT on a file?
pip install moshi, then python -m moshi.run_inference --hf-repo kyutai/stt-2.6b-en audio.mp3. For production streaming, use the moshi-server Rust crate; on Apple devices, the MLX implementation.
What licence is Kyutai STT under?
The model weights are CC-BY-4.0; the code in the delayed-streams-modeling repository is Apache-2.0.
What is Kyutai STT?
Kyutai STT streams transcripts as people speak: English and French (1B, 0.5 s delay) or English (2.6B, 2.5 s delay), word timestamps, CC-BY-4.0. Rust server for production, MLX for Apple devices.
Is Kyutai STT free?
Yes. Kyutai STT is free and open source (free (CC-BY-4.0 weights)).
Does Kyutai STT work on Mac, Windows and Linux?
Kyutai STT runs on Linux, macOS and iOS.
How do I install Kyutai STT?
With pip: `pip install moshi # python -m moshi.run_inference --hf-repo kyutai/stt-2.6b-en audio.mp3`. You need Python and, for most audio tools, ffmpeg on your PATH first.
How many languages does Kyutai STT support?
Kyutai STT lists 2 languages.
Does Kyutai STT give word-level timestamps?
Yes — Kyutai STT produces timing per word, which is what subtitle cues and clip boundaries need.
Can Kyutai STT transcribe live audio?
Yes — Kyutai STT supports streaming / live input, not just finished files.
Is Kyutai STT HIPAA compliant?
Kyutai STT does not claim HIPAA compliance. For PHI, run it on infrastructure you control or choose a service that will sign a BAA.
What are the limitations of Kyutai STT?
English and French only (2.6B is English-only). Tuned for real-time streams rather than file batches, and serving it means running the Rust server on a GPU.
Who is Kyutai STT best for?
Builders of voice agents and live captioning in English or French who need low-latency streaming and batching on GPUs.
What should I know before choosing Kyutai STT?
From the team behind Moshi. For recordings that already exist, a file transcriber is the simpler tool.
Whipscribe is a hosted transcription service. If you want transcripts without running infrastructure, paste a URL or drop a file in the form below — you'll have a transcript in seconds.
Transcribe your recordings
Running Kyutai STT yourself? Send the same file here and compare the transcript. Transcribe any language, any recording · Transcription services from $0.29 an hour · Transcribe from any meeting or recording platform · How to transcribe, step by step
Explore
All transcription tools · By language · By platform · By file format · How-to guides · Transcription services · Transcription API · Pricing
Have the recording already?
Skip the setup and send the file or the link: see transcribing a long MP3 or M4A, transcribing a podcast episode and an SRT subtitle file from any recording.
Rather have it done for you?
Free tools cost setup time. When you want a finished transcript instead: see an honest roundup of the best transcription services, affordable transcription priced per hour and transcription services from $6.