Voice Cloning on LibriSpeech-PC clean (test)
2.05WERCosyVoice 2
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CosyVoice 2Training data=167K Multi., Params=618M, Zero-shot=true2026.06 | 2.05 | 65.9 | 4.38 | |
| Ground TruthZero-shot=true2026.06 | 2.48 | 69.5 | 4.1 | |
| FireRedTTSTraining data=248K Multi., Params=~580M, Zero-shot=true2026.06 | 2.69 | 47 | — | |
| BareWave (proposed training scheme)Training data=19.4k EN, Params=983.6M, Zero-shot=true2026.06 | 2.88 | 61.4 | 4.01 | |
| F5-TTS Base (+ Vocos)Training data=19.4k EN, Params=335.8M + 13.5M, Zero-shot=true2026.06 | 3.17 | 59.7 | 4.1 | |
| BareWave (basic training)Training data=19.4k EN, Params=983.6M, Zero-shot=true2026.06 | 3.32 | 47.1 | 3.69 | |
| Simple direct-wave baselineTraining data=19.4k EN, Params=983.6M, Zero-shot=true2026.06 | 3.45 | 41.6 | 3.6 | |
| E2-TTS Base (+ Vocos)Training data=19.4k EN, Params=333M + 13.5M, Zero-shot=true2026.06 | 4.32 | 63.2 | 3.84 |