Issues with this dataset
#3
by
mkarapka - opened
I’ve noticed several issues with the dataset. Some TTS audios are several minutes long, where about 95% of the duration is either silence or random noise – for example in the mls-tts-fish_speech configuration, key = dev/1093_0. Additionally, the key field could use a more convenient format, and it would be helpful if corresponding speaker_ids from the original dataset were preserved.