Timestamps:
00:00 - Intro
01:07 - Pre Reqs
02:14 - Local Install
04:59 - Initial Testing
06:34 - VRAM Test
07:51 - Voice Cloning Explanation
10:15 - Voice Clone Testing
12:07 - Release Drama
12:35 - More Testing
15:52 - Cloning My Voice
19:37 - Closing Thoughts
The Sesame CSM 1B is an open-source speech model that brings lifelike AI-generated voices to local hardware. Originally making waves with its impressive online demo, Sesame has now released its first open-source voice model, allowing for efficient and powerful voice generation entirely offline.
In this video, we take a first look at the Sesame CSM 1B model, exploring its capabilities and installing it locally on a laptop with an RTX 4060 and just 8GB of VRAM. We walk through the installation process step-by-step, run initial speech generation tests, and evaluate its VRAM usage and efficiency. Then, we dive into its voice cloning capabilities, testing how well it can replicate different voices—including cloning my own.
We also discuss some of the drama surrounding its release, before finishing with a final test and analysis of its real-world performance. This video provides a detailed look at what makes Sesame CSM 1B an exciting local speech model for AI voice applications.
Repo Link: https://github.com/SesameAILabs/csm