Modelpublic

anak10thn/jalak-vits-multispeaker-slr44

sourceHugging Facecc-by-sa-4.0updated 3mo agoView on Hugging Face

0likes40downloads

12 commits on main

080d8cb3mo ago

FIX: use g2p-id (Wikidepia, the ORIGINAL dataset phonemizer) instead of espeak-id. Output has ʔ + per-word stress matching training data. 25.3s sample.

anak10thn

bb85b803mo ago

Add diagnostic samples: pretrained vs finetuned, IPA-asli vs espeak-id

anak10thn

e310dcf3mo ago

Add diagnostic samples: pretrained vs finetuned, IPA-asli vs espeak-id

anak10thn

2114c633mo ago

Add diagnostic samples: pretrained vs finetuned, IPA-asli vs espeak-id

anak10thn

7c7e7e23mo ago

Add diagnostic samples: pretrained vs finetuned, IPA-asli vs espeak-id

anak10thn

d36266b3mo ago

Add diagnostic samples: pretrained vs finetuned, IPA-asli vs espeak-id

anak10thn

bb801393mo ago

Add diagnostic samples: pretrained vs finetuned, IPA-asli vs espeak-id

anak10thn

b30eb1b3mo ago

Add diagnostic: generate from dataset-native IPA vs espeak-id (phonemizer mismatch test)

anak10thn

9658b1d3mo ago

Add diagnostic: generate from dataset-native IPA vs espeak-id (phonemizer mismatch test)

anak10thn

87cce793mo ago

Add long audio sample (42.5s, speaker wibowo)

anak10thn

cc4d6123mo ago

Add jalak VITS finetuned model (124 spk, +SLR44 Sundanese, eval_mel=21.19)

anak10thn

4850b893mo ago

initial commit

anak10thn