VoxCPM2 GitHub Repository
VoxCPM2: Tokenizer-Free TTS for Multilingual Speech Generation, Creative Voice Design, and True-to-Life Cloning.
Benchmark Data (Minimax-MLS-test WER)
- CosyVoice3-1.5B: 3.91 (EN) / various language scores
- Fish Audio S2: 2.65 (EN)
- VoxCPM2: 3.65 (EN) / 5.00 / 8.55 / 8.48 / 5.96 / 5.69 / 4.77 / 3.80 / 9.85 / 4.25 / 5.21
- Qwen3-TTS: mentioned as competitor in benchmark