Tools/Chatterbox
Chatterbox
Resemble AI's open-source (MIT-licensed) TTS/voice-cloning model family — zero-shot cloning from a ~5-10s reference clip, emotion/exaggeration control, and built-in Perth watermarking. Three variants: Turbo (350M, low-latency, supports paralinguistic tags like [chuckle]/[sigh]), Original (500M, emotion control), Multilingual (500M, 23+ languages). The MIT-licensed, actively-maintained successor that replaced VoiceBox/XTTS-v2 in the local voice-cloning workflow.
See TechResearch/Chatterbox_Experiment/chatterbox_experiment for the full deep-research writeup, and TechResearch/Chatterbox_Experiment/chatterbox_implementation_log for the hands-on Apple Silicon self-hosting build log.