Long Speech Generation on Curated internal dataset long-en
1.225WERQwen3-TTS-25Hz-1.7B-CustomVoice
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen3-TTS-25Hz-1.7B-CustomVoiceFrequency=25Hz, Model Size=1.7B, Voice Type=CustomVoice2026.01 | 1.225 | |
| VibeVoice2026.01 | 1.78 | |
| Qwen3-TTS-12Hz-1.7B-CustomVoiceFrequency=12Hz, Model Size=1.7B, Voice Type=CustomVoice2026.01 | 2.812 | |
| MiniCPM-o 4.52026.04 | 3.37 | |
| Higgs-Audio-v2Synthesis Architecture=chunk-based2026.01 | 6.917 | |
| VoxCPM2026.01 | 7.474 | |
| CosyVoice22026.04 | 14.8 | |
| Qwen3-Omni2026.04 | 17.33 |