# Moshi: a speech-text foundation model for real-time dialogue

*Source* · `src-defossez-2024-moshi` · status: **n/a**

| field | value |
|---|---|
| kind | Preprint |
| authors | Alexandre Défossez, Laurent Mazaré, Manu Orsini, Amélie Royer, Patrick Pérez, Hervé Jégou, Edouard Grave, Neil Zeghidour |
| venue | arXiv |
| date | 2024-09-17 |
| url | https://arxiv.org/abs/2410.00037 |
| arxiv | 2410.00037 |
| accessed | 2026-10-24 |

## Notes

Full-duplex spoken dialogue as speech-to-speech generation from a text LM backbone; models user and system speech as parallel codec token streams; Inner Monologue predicts time-aligned text before audio. Latency 160 ms theoretical, 200 ms practical.

## Backlinks (computed)

**cited by:** [Realtime speech-to-speech and expressive TTS for characters](../technique/realtime-voice-models.md)

## Citation

Alexandre Défossez, Laurent Mazaré, Manu Orsini, Amélie Royer, Patrick Pérez, Hervé Jégou, Edouard Grave, Neil Zeghidour (2024-09-17). *Moshi: a speech-text foundation model for real-time dialogue*. arXiv https://arxiv.org/abs/2410.00037

---
Provenance: created in run `2026-10-24-techniques-llm-agents`.