VoxCPM2 Needs Just 8GB VRAM to Clone and Generate Any Voice Locally
TL;DR
VoxCPM2, developed by OpenBMBB, introduces a self-hosted approach to text-to-speech (TTS) technology, combining voice generation, cloning and multilingual synthesis into one cohesive system. Unlike cloud-based alternatives, this local AI model prioritizes data privacy and operational control, making it particularly appealing for industries with strict confidentiality requirements.
Nauti's Take
The first useful test is a local run with approved voice data, measuring VRAM usage, latency, and intelligibility under realistic load. The 8 GB figure is reported by a single outlet here, so teams should verify it against the official model, licensing terms, and voice cloning rules before considering production use.
Video
Sources