Voice Clone Tutorial Scripts
Added 2023-05-08 20:48:05 +0000 UTCTutorial published : https://youtu.be/OiMRlqcgDL0
GitHub file link that has many instructions (the one used in the video) : https://github.com/FurkanGozukara/Stable-Diffusion/blob/main/Tutorials/Deep-Voice-Clone-Tutorial-Tortoise-TTS.md
pre_process_given_text_file. py will be used by process_given_speech_text.bat automatically to preprocess and fix prepare your speech text file.
process_given_speech_text.bat will automate your process. just drag and drop your speech txt file into it.
epoch_comparison.bat will automate your process and will generate speech files from given text file by using all of the checkpoint files you put in the given folder. it also works with drag and drop the text file
put both files into the same folder. Don't forget to modify necessary variables as explained in the GitHub repo and in the tutorial video
setting up the necessary variables are mandatory in the given files so check out video or GitHub page
fixDuration.py this is a new file. With this file you can remove bigger than any length sounds. For example it will remove above 12 seconds right now. Run this script inside the training set folder which have wavs folder, train.txt and valid.txt files. Moreover in ozen toolkit, you can change config.ini file and set seg_min_duration = 5.0 for removing smaller than 5 second speech files automatically. But this will also make you lose a lot of speech parts. You can change valid_ratio , seg_onset , and seg_offset values in ozen toolkit config.ini to see what kind of effects they are making when pre processing given speech audio file.
Comments
yes there is a way to resume. in the gui you give a path of pth. you can give your last checkpoint. by the way coqui is much faster and better. for 150$ private session, i can share my prepared scripts and show you how to voice clone training. i will hopefully make a tutorial but not ready yet.
Furkan Gözükara
2023-12-21 16:40:12 +0000 UTCOk- so my computer stopped after about 6 weeks of training because of a power outage. It got to 11400_gt.pth and 11400.state - is there a way to restart it or go with what I already have?
Jessica
2023-12-21 15:34:33 +0000 UTCI think it doesn't affect a lot. I also had same problem when training
Furkan Gözükara
2023-10-16 15:54:55 +0000 UTCNevermind - I did your steps in order without trying to start DLAS and it is working so far :D Is this user warning fatal though? 38: UserWarning: Detected call of `lr_scheduler.step()` before `optimizer.step()`. In PyTorch 1.1.0 and later, you should call them in the opposite order: `optimizer.step()` before `lr_scheduler.step()`. Failure to do this will result in PyTorch skipping the first value of the learning rate schedule. See more details at https://pytorch.org/docs/stable/optim.html#how-to-adjust-learning-rate warnings.warn("Detected call of `lr_scheduler.step()` before `optimizer.step()`. "
Jessica
2023-10-16 02:45:29 +0000 UTCI see. yes i have modified it. now when you run process_given_speech_text.bat it will read lines and merge them with ; . so now pre_process_given_text_file.py split speech text into lines
Furkan Gözükara
2023-06-01 22:50:25 +0000 UTCI was only trying to run pre_process_given_text_file.py standalone to check the result but there are no separators in the file.
2023-06-01 21:12:39 +0000 UTCNow the other file is merging lines with ; when generating speech. this change was made to support special characters like ! I spent quite a bit time to make it work. Can you try and let me know. Thank you for patron support
Furkan Gözükara
2023-06-01 20:35:34 +0000 UTC
