metadata

pretty_name: DNSMOS Score for common TTS datasets

(Placeholder)

DNSMOS-TTS

DNSMOS-TTS contains DNSMOS Scores for common TTS datasets

This repo uses Lhotse to manage datasets.

For example, to load LJ-Speech:

from lhotse import CutSet

for cut in CutSet.from_webdataset("pipe:curl -s -L https://huggingface.co/datasets/Gatozu35/DNSMOS-TTS/resolve/main/ljspeech_mos.tar"):
    wav = cut.load_audio()
    mos = cut.supervisions[0].custom["mos"]
    ...

If you don't want to use lhotse, I have also uploaded a csv of the scores for each id.