KyutaiKyutai STT
Kyutai STT 2.6B English
Kyutai STT 2.6B English transcribes incoming speech while audio is still arriving. It uses the Delayed Streams Modeling approach and produces text with a configured delay instead of waiting for a complete recording.
Research release
The model is designed for a live audio stream
The larger release is English-only. Its longer configured delay gives the model more incoming speech before it emits aligned text. Its semantic voice activity signal helps applications identify speech activity. Candle, MLX and Transformers packages are deployment forms of the same release.stt-2.6b-en: Kyutai model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | kyutai/stt-2.6b-enkyutai/stt-2.6b-en-mlxkyutai/stt-2.6b-en-candlekyutai/stt-2.6b-en-trfs |
| Languages | English.stt-2.6b-en: Kyutai model card |
| Model delay | 2.5 seconds; this is not a full application response-time guarantee.stt-2.6b-en: Kyutai model card |
| Weight license | CC BY 4.0.stt-2.6b-en: Kyutai model card |
Related models
- Kyutai STT 1B English/French
Kyutai ยท Kyutai STT
A streaming English and French transcription model with a configured output delay of 0.5 seconds.
Launch date not confirmedResearch releaseDownloadable