ResearchBenchmarksText-to-Speech on LJSpeech (WER, MCD, Log-F0 RMSE, UTMOSv2)Follow3.37WEROne-shot3.19844.35675.5156.6733Mar 14, 2026Evaluation ResultsMethodMethodLinksWERMCDLog-F0 RMSEUTMOSv2One-shot2026.033.375.9140.3324.05Grad-TTS2026.034.385.8720.334.024UDDDecoding Strategy=SampleDecoding Strategy=Sample2026.034.555.860.3443.989UDDDecoding Strategy=ArgmaxDecoding Strategy=Argmax2026.034.715.830.3324.003TDDDecoding Strategy=SampleDecoding Strategy=Sample2026.036.316.0570.3413.867TDDDecoding Strategy=ArgmaxDecoding Strategy=Argmax2026.037.665.850.3413.988