Sesame CSM

Sesame CSM is a speech generation model for developers and researchers building voice applications on their own hardware. It generates spoken output from text and audio inputs, including earlier parts of a conversation.

• Contextual speech: You can provide previous utterances and their audio to guide generation; the project recommends context for better results.
• Multiple speakers: Speaker-specific prompts support generated dialogue between characters, while generation without a prompt uses a random speaker identity.
• Application integration: CSM works with Hugging Face Transformers and provides Python code for using the model in your own applications.

The released checkpoint is CSM-1B. Local execution requires a CUDA-compatible GPU and access to CSM-1B and Llama-3.2-1B on Hugging Face. The Python implementation supports Windows. The repository code uses Apache 2.0.

A separate hosted Hugging Face Space lets you test audio generation. Sesame's interactive voice demo uses a fine-tuned CSM variant, so the public checkpoint isn't the exact model behind that demo.