You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Requested in #47 (comment) . Looking for someone to add a train/validation/test split feature. This is something we can use for the encoder, synthesizer and vocoder.
Here is an example for how it could work, using encoder as an example:
Add new train, dev, and test folders under datasets_root/SV2TTS/encoder. Each of the 3 new folders has identical structure to the what is currently inside datasets_root/SV2TTS/encoder.
Define new hparams called train_split, dev_split and test_split. This is the proportion of the data to be assigned to each category.
Preprocessing randomly assigns speakers to each category and writes the output files to the corresponding folder.
During training, the model is evaluated using speakers from the dev or test sets at predefined intervals.
The text was updated successfully, but these errors were encountered:
Requested in #47 (comment) . Looking for someone to add a train/validation/test split feature. This is something we can use for the encoder, synthesizer and vocoder.
Here is an example for how it could work, using encoder as an example:
train
,dev
, andtest
folders underdatasets_root/SV2TTS/encoder
. Each of the 3 new folders has identical structure to the what is currently insidedatasets_root/SV2TTS/encoder
.train_split
,dev_split
andtest_split
. This is the proportion of the data to be assigned to each category.The text was updated successfully, but these errors were encountered: