i mean at least most of them are short clips :]

Yeah

it'd be much worse if they were all around 10 seconds or more

yep

Good thing they are about 1 to 3 seconds long

and they are all with no echo

So the model will sound nicely

i'm really gonna pay for colab pro just so i can hear my model

colab pro?

it says i cant connect to a gpu

due to "usage limits"

Maybe you tried to reset the thing?

Left it on for too long, most likely.

oh well

I got the same message after I trained a model for 6 hours straight.

at least now i'll be able to do a higher batch size model

god damn its still just noises

does it sound different on the site?

did i not train the model long enough?

i got it to say someones i guess

that is the closest i got to anything

hey Im collecting data and wondering whats the structure of the wav file transcript file should i use? i have been making wave files of short clips 5-30s but i can get hundreds of them if needed. Then a corresponding text file with the same name but .txt instead of .wav. Any tips for removing background noise/music? I have been skipping over collecting data when people are talking over each other any tools for separating voices its not really worth it just curious? I have just collected 50 of these snippets and wanted to make a baseline. any feedback is appreciated thanks

for the transcript, it should be like this:

wavs/filename.wav|This is where the transcript goes.

throughout the file, one file per line with no blank lines. also your clips shouldn't be more than 13-15 seconds long, so you'll have to trim some of them down. as for removing noise/music, there's plenty of options, such as AI powered ones like lalal.ai or vocalremover.org, a Colab notebook, etc.

Just i need to relax and search spanish voice clips.

thanks i looked at the other notebook for a sec need to make a better data pipeline anyway just wanted to see a rough model. what does the "|" it sounds like I'm doing it right with one transcript per file no newlines. I should just put everything in the wav/ folder? fom the notebook it looks like it just wants one text file?

Audio clips go to the wavs folder during training, transcript goes to filelist folder.

yeah, put all your audio in the wav folder (and make sure it's all mono and has a 22050 sample rate)

the transcript file should be outside of the wav folder

so it should be one file?

the transcript, yes