CSM

Sesame’s 1B conversational speech model behind the viral Maya voice demo.

Install Repository Compare

Fit.

Catalog snapshot

Platforms: Linux, Macos, Windows
Operating systems: Linux, macOS, Windows
Ram floor: 8 GB
Gpu: Optional · 6 GB VRAM helps
Docker: No
Setup: Intermediate
Deploy as: Local
Commercial use: Allowed

Weights are Apache-2.0; the repo is a research drop, not a product.

Project.

Fetched Sep 12, 2026¹

License: Apache 2.0
Stars: 14.7k¹
Forks: 1.5k¹
Open issues: 9¹
Language: Python¹
Last commit: May 27, 2025¹
Tags: ttsresearchllama

About.

CSM-1B is Sesame’s conversational speech model: a Llama backbone that generates audio tokens, with listening and speaking checkpoints for the famous Maya demo. Research-grade code, dormant since mid-2025.

1B Llama backbone · conversational context · Apache-2.0 weights · studio demo

Alternatives.

Nearby.

All Speech AI

WWhisperOpenAI’s speech-to-text model you can run locally.Compare
PPiperFast local neural text-to-speech that runs on CPU.Compare
Wwhisper.cppC/C++ port of Whisper for fast local transcription.Compare
FFaster-WhisperWhisper reimplementation on CTranslate2 for faster local ASR.Compare
WwhisperXWhisper with word-level timestamps and speaker diarization for real subtitles.Compare