A speech-input LLM that feeds Whisper encoder features through a learned projector into Llama 3.2 1B Instruct, so it can respond to spoken input without a separate transcription step. It replies in text, so pair it with a TTS voice.
kyutai/stt-1b-en_fr
Indexed from Hugging FaceLinked to its home on Hugging Face. FrontierOS indexes it; nothing is hosted or mirrored here.ModelSpeech to text989M paramsEst. Edge GPUEstimated minimum compute for inference: Jetson Orin class, 8–16 GB shared memory. Not a benchmark.CC BY 4.0
About
A decoder-only streaming English and French transcriber based on Kyutai's delayed-streams design, running about 0.5 s behind the audio and exposing a semantic voice-activity signal for end-of-turn detection.
Why it matters for physical AI
Low-latency streaming STT with end-of-turn detection helps conversational robots avoid both interrupting people and lagging behind them.
Summary written by FrontierOS in our own words. The upstream card and license govern use. Metadata checked 27 Sep 2026. Read the full card on Hugging Face
Files, weights and version history live on Hugging Face. FrontierOS indexes this repo; it does not host or mirror it.
hf download kyutai/stt-1b-en_frUses the Hugging Face CLI (pip install -U huggingface_hub). This downloads the whole repo; add --include "<pattern>" to fetch only the files you need.
Full-duplex or cascaded voice for robots?
Full-duplex models such as Moshi can listen while they talk, which makes interruptions and backchannels feel natural. Cascaded stacks are easier to debug and swap piece by piece.
More speech to text
Browse allA faster Whisper large-v3 derivative whose decoder is cut from 32 layers to 4, giving multilingual transcription in roughly 100 languages at a small accuracy cost (it was not fine-tuned for translation).
The small size of the Moonshine Streaming family, an English-only encoder-decoder recognizer that uses a sliding-window encoder so it can transcribe live audio with low latency on edge hardware.