CSM
Sesame's open-weight conversational speech model that turns text into natural, context-aware dialogue audio.
About
CSM (Conversational Speech Model) is the 1-billion-parameter speech model Sesame open-sourced in March 2025 under Apache 2.0. It generates audio from text plus prior conversation audio, using a Llama backbone with a small audio decoder, and it sounds best when given dialogue context so it can match tone and pacing. It is a speech generator, not a chatbot: it cannot produce text on its own, so you pair it with a separate language model. It is English-only and released for research and education, and the impressively human Maya/Miles voice demo on Sesame's site runs a fine-tuned variant that was never released.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (verified tier, metrics-v1). Evidence: Apache 2.0 confirmed in the HF repo metadata (license: apache-2.0); Released 2025-03-13; native in Hugging Face Transformers since 4.52.1 (2025-05-20); Official FAQ: cannot generate text, English only, base model ships no specific voices; Released for research and educational purposes with explicit misuse prohibitions (impersonation, fraud, misinformation); Maya/Miles demo is an unreleased fine-tuned variant, not the open checkpoint; No provider updates since the May 2025 Transformers integration; the repo is effectively static; HuggingFace access is gated: you must accept the license before download. License: Apache 2.0 (confirmed in HF repo metadata); weights access is gated behind license acceptance..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 8.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · Apache 2.0 confirmed in the HF repo metadata (license: apache-2.0)
- · Released 2025-03-13; native in Hugging Face Transformers since 4.52.1 (2025-05-20)
- · Official FAQ: cannot generate text, English only, base model ships no specific voices
- · Released for research and educational purposes with explicit misuse prohibitions (impersonation, fraud, misinformation)
- · Maya/Miles demo is an unreleased fine-tuned variant, not the open checkpoint
- · No provider updates since the May 2025 Transformers integration; the repo is effectively static
- · HuggingFace access is gated: you must accept the license before download