Text-to-Speech on VCTK 5H training budget v0.92
8.5WER (%)K=7 SPARCLE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| K=7 SPARCLEInput representation=SPARCLE, Number of unfrozen layers (K)=7, Timbre embeddings=with (+T), TTS Backend=ParrotTTS2026.05 | 8.5 | 0.98 | |
| Unfrozen SPARCLEInput representation=SPARCLE, Timbre embeddings=with (+T), TTS Backend=ParrotTTS, Adaptation=Fully unfrozen2026.05 | 8.9 | 0.98 | |
| K=3 SPARCLEInput representation=SPARCLE, Number of unfrozen layers (K)=3, Timbre embeddings=with (+T), TTS Backend=ParrotTTS2026.05 | 10.2 | 1.18 | |
| K=7 SPARCLEInput representation=SPARCLE, Number of unfrozen layers (K)=7, Timbre embeddings=without (-T), TTS Backend=ParrotTTS2026.05 | 11 | 1.21 | |
| K=3 SPARCLEInput representation=SPARCLE, Number of unfrozen layers (K)=3, Timbre embeddings=without (-T), TTS Backend=ParrotTTS2026.05 | 11.7 | 0.98 | |
| Unfrozen SPARCLEInput representation=SPARCLE, Timbre embeddings=without (-T), TTS Backend=ParrotTTS, Adaptation=Fully unfrozen2026.05 | 11.9 | 0.97 | |
| K=1 SPARCLEInput representation=SPARCLE, Number of unfrozen layers (K)=1, Timbre embeddings=without (-T), TTS Backend=ParrotTTS2026.05 | 12.2 | 1 | |
| K=1 SPARCLEInput representation=SPARCLE, Number of unfrozen layers (K)=1, Timbre embeddings=with (+T), TTS Backend=ParrotTTS2026.05 | 12.6 | 0.98 | |
| Frozen SPARCLEInput representation=SPARCLE, Timbre embeddings=without (-T), TTS Backend=ParrotTTS, Adaptation=Frozen2026.05 | 13.2 | 1.33 | |
| Frozen SPARCLEInput representation=SPARCLE, Timbre embeddings=with (+T), TTS Backend=ParrotTTS, Adaptation=Frozen2026.05 | 13.3 | 1.33 | |
| Char EmbeddingsInput representation=Character, Timbre embeddings=without (-T), TTS Backend=ParrotTTS2026.05 | 17.2 | 1.37 | |
| Phone EmbeddingsInput representation=Phoneme, Timbre embeddings=without (-T), TTS Backend=ParrotTTS2026.05 | 27.7 | 1.37 |