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.
For a robot that also has to act, which would you choose today, and why?
Referenced repos
Indexed from Hugging Face and linked upstream, not hosted here.
Kyutai's full-duplex speech-to-speech foundation model in its male-voice variant, which models the user's and its own speech as parallel streams and reaches about 200 ms latency in practice.
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.
A mixture-of-experts omni model with about 3B active parameters that takes text, image, audio, and video as input and replies in both text and streaming speech.
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.