The first set was trained for 441K steps on theLJ Speech Dataset Speech started to become intelligible around 20K steps. The second set was trained by@MXGrayfor 140K steps on theNancy Corpus. @npuichigofixeda bug where dropout was not being applied in the prenet. ...
Here is the expected loss curve when training on LJ Speech with the default hyperparameters: If you used mimic-recording-studio and want to create an ljspeech dataset syntax out of it you can use the following command python3 ./datasets/createljspeech.py --mrs_dir=<path_to>/mimic-record...
Here is the expected loss curve when training on LJ Speech with the default hyperparameters: If you used mimic-recording-studio and want to create an ljspeech dataset syntax out of it you can use the following command python3 ./datasets/createljspeech.py --mrs_dir=<path_to>/mimic-record...
Here is the expected loss curve when training on LJ Speech with the default hyperparameters: If you used mimic-recording-studio and want to create an ljspeech dataset syntax out of it you can use the following command python3 ./datasets/createljspeech.py --mrs_dir=<path_to>/mimic-record...
Here is the expected loss curve when training on LJ Speech with the default hyperparameters: If you used mimic-recording-studio and want to create an ljspeech dataset syntax out of it you can use the following command python3 ./datasets/createljspeech.py --mrs_dir=<path_to>/mimic-record...
Here is the expected loss curve when training on LJ Speech with the default hyperparameters: If you used mimic-recording-studio and want to create an ljspeech dataset syntax out of it you can use the following command python3 ./datasets/createljspeech.py --mrs_dir=<path_to>/mimic-recordin...
Here is the expected loss curve when training on LJ Speech with the default hyperparameters: If you used mimic-recording-studio and want to create an ljspeech dataset syntax out of it you can use the following command python3 ./datasets/createljspeech.py --mrs_dir=<path_to>/mimic-recordin...
Here is the expected loss curve when training on LJ Speech with the default hyperparameters: If you used mimic-recording-studio and want to create an ljspeech dataset syntax out of it you can use the following command python3 ./datasets/createljspeech.py --mrs_dir=<path_to>/mimic-record...
Here is the expected loss curve when training on LJ Speech with the default hyperparameters: If you used mimic-recording-studio and want to create an ljspeech dataset syntax out of it you can use the following command python3 ./datasets/createljspeech.py --mrs_dir=<path_to>/mimic-record...
The first set was trained for 441K steps on theLJ Speech Dataset Speech started to become intelligible around 20K steps. The second set was trained by@MXGrayfor 140K steps on theNancy Corpus. @npuichigofixeda bug where dropout was not being applied in the prenet. ...