List your company

Zero-shot Text-to-Speech on LibriSpeech LS960 (test-clean)

1.6WER (Whisper-large)Ground truth
1.4642.3823.34.218May 28, 2026

Evaluation results

MethodLinks
Ground truthText=-, Speech=-, Freq=-2026.051.62.20.925
DACText=-, Speech=-, Freq=50.02026.051.62.20.922
HiFi-GanText=-, Speech=-, Freq=62.52026.051.62.20.903
MELDText=BPE, Speech=Mel, Freq=62.52026.051.92.40.872
MELDText=BPE, Speech=Mel, Freq=31.32026.051.92.50.855
Codec-LMText=Phn, Speech=DAC, Freq=50.02026.054.75.70.872
Codec-LMText=BPE, Speech=DAC, Freq=50.02026.054.85.30.864
VALL-EText=Phn, Speech=Encodec, Freq=75.02026.055—0.868