Conversational Speech Model (CSM) and Maya/Miles demo
Sesame's CSM voice demo crossed the 'uncanny valley' for many listeners; CSM-1B was open-sourced under Apache 2.0.
A two-transformer model that generates audio codes conditioned on conversation history, producing natural prosody and breaths. Demo went viral in March 2025; 1B-parameter weights on Hugging Face (repo created 2025-03-06).
- Date
- Thursday, 27 February 2025
- Lab
- Sesame
- Kind
- model
- Access
- open weights
Blog dated 2025-02-27; the architecture description and 'viral' framing are from memory beyond the post's opening paragraphs.
Sources
This record was checked against its sources on 6 October 2026. How we check