AIToolScan

VoxCPM - Open-Source Tokenizer-Free Text-to-Speech

VoxCPM — tokenizer-free text-to-speech with creative voice design and cloning

VoxCPM is an open-source, tokenizer-free text-to-speech system from OpenBMB (ModelBest). It directly generates continuous speech representations through an end-to-end diffusion autoregressive architecture, bypassing discrete tokenization for natural, expressive synthesis. The project ships a family of models and is released under the Apache-2.0 license for commercial use.

Model family

  • VoxCPM2 (2B, latest): 30 languages, Voice Design, Controllable Voice Cloning, 48kHz studio-quality audio.
  • VoxCPM1.5 (0.8B): Chinese and English, continuation cloning, 44.1kHz audio.
  • VoxCPM-0.5B (0.6B): Chinese and English, 16kHz audio.

Highlights

  • Voice Design: create a new voice from a natural-language description, no reference audio required.
  • Controllable and ultimate voice cloning with style guidance.
  • Context-aware prosody and real-time streaming (RTF ~0.3 on RTX 4090).
  • Production serving via Nano-vLLM and vLLM-Omni; on-device inference via llama.cpp-omni.

Links