Check if the notebook you're using is the one from #841437191073955920 , if not, try using it

thanks, but im getting another issue

126 dirs = [x for x in os.scandir(target_path)]
    127 if len(dirs) == 1:
--> 128   with open(sorted(glob.glob(os.path.join(dirs[0].path, "*.txt")))[0]) as f:
    129     already_multispeaker = (len(f.readline().split("|")) == 3)
    130 
IndexError: list index out of range

might be due to the fact that i wrote "transcripts" instead of "transcription"

make sure your txt has no blank lines in it

hmm still getting the same issue

send the txt here

It's missing the wavs location

does tacotron have its own formatting

i used whisper for this extract

yes

(wav location)|(text)

so, something like

swag

wavs/01.wav|Why are you talking to us yes I am.

Also, there's a bunch of lines with one word, you'll have to clean that by hand

One word won't be enough for one audio

is the format correct?

lol should be .wav

cool, still getting the same error

try adding punctuation at the end of each line

is it related to format?

YEAH I dunno if pipeline does it but legacy always threw a fit when we didn't close a line with at LEAST a period so your first line should prolly be smth like

wavs/cooltext.wav|Maybe now.

and you can use most punctuation, I've gotten away with having questions and exclamations and it read em just fine

hmmmmmmmmmmmmmmmmmmm still scuffed

maybe i'll just use tcotron's transcribe

any cool tips on how to ensure quality of data model? how important is data cleaning? should i make sure non human voice is removed? any way i can avoid paying premium on google colab while transcribing?

Clean data is very important, especially with dataset, i wanna say 20 minutes total or less? Bigger ones can handle more, like with some whistles or just noise that couldn't be removed in general by cleaners. Effects can be fine as long as itd not too harsh. There's also the Whisper NB for transcripts which does a decent job for what I've thrown at it.

I have used the large model though so idk how the smaller ones fair

You have only 1 wav?

each line must point to a different file which corresponds to the transcription after the "|"