OpenBMB Releases VoxCPM2: Tokenizer-Free Multilingual Speech Synthesis

VoxCPM2 is a Tokenizer-Free TTS model that models raw audio directly, avoiding the information loss caused by traditional tokenizers. It supports multilingual speech generation, creative voice design, and realistic voice cloning.

It covers 30 languages and 8 major dialects, outputting 48kHz studio-grade audio. The GitHub repository is actively maintained with positive community feedback.

The model represents an important direction for TTS architecture evolution — moving away from dependence on discrete tokenizers to achieve more natural speech synthesis at higher sampling rates.