List your company

Cross-sentence Zero-shot Speech Synthesis on LibriSpeech clean (test)

1.9WERHVoicebox
1.623.515.47.29Jul 11, 2024

Evaluation results

MethodLinks
VoiceboxTraining Data Hours=60K2024.071.9—66.2
MELLETraining Data Hours=50K2024.072.11.4762.5
MELLE-R2Training Data Hours=50K, reduction factor=22024.072.141.560.8
Ground Truth2024.072.151.6177.9
MELLE-R3Training Data Hours=50K, reduction factor=32024.072.191.5157
Ground Truth (mel-spectrogram)2024.072.241.6473.2
MELLE-R4Training Data Hours=50K, reduction factor=42024.072.31.5653.2
Ground Truth (EnCodec, 8 codebooks)2024.072.331.6571
VALL-E 2Training Data Hours=50K2024.072.441.564.3
MELLE-R5Training Data Hours=50K, reduction factor=52024.072.721.9650.6
RALL-ETraining Data Hours=44K2024.072.82.549
MELLE-limitedTraining Data Hours=9602024.072.82.2159.1
VALL-E RTraining Data Hours=9602024.073.973.1836.5
CLaM-TTSTraining Data Hours=55K2024.075.11—49.5
VALL-ETraining Data Hours=60K2024.075.9—58
ELLA-VTraining Data Hours=9602024.078.97.1530.7