# machine-learning
- u
user
04/14/2021, 1:08 AMno - u
user
04/14/2021, 1:08 AMyou can use 20 - t
Toasty
04/14/2021, 1:08 AM20 will work? - u
user
04/14/2021, 1:08 AMyes - t
Toasty
04/14/2021, 1:08 AMAlright - t
Toasty
04/14/2021, 1:08 AMMy theory is that I made the files too big - t
Toasty
04/14/2021, 1:09 AMAnd the only way to counteract is to lower the batch size - t
Toasty
04/14/2021, 1:14 AMLittle did I realize, I made my dataset untrainable. - t
Toasty
04/14/2021, 1:16 AMThree days of work gone. - t
Toasty
04/14/2021, 1:17 AMfml. - u
user
04/14/2021, 1:17 AMFuck google - t
Toasty
04/14/2021, 1:18 AMThe only way to make my dataset from 24 minutes to 2 hours was raise the limit on ASSFAP from 12 seconds to 30. - t
Toasty
04/14/2021, 1:18 AM(yes ik thats the actual name) - t
Toasty
04/14/2021, 1:18 AMI feel like I made my dataset too big for Tacotron2. - u
user
04/14/2021, 1:18 AMRip - t
Toasty
04/14/2021, 1:19 AMSo... it means I'm stuck with a crap 24 minute tts of Yahtzee - t
Toasty
04/14/2021, 1:20 AMcreated with an automated data set maker that's crap - u
user
04/14/2021, 1:20 AMWich one? - t
Toasty
04/14/2021, 1:21 AMAutomatic Super Speaker-Filtered Audio Processing V4 - u
user
04/14/2021, 1:21 AMAnd does it build wav files in? - t
Toasty
04/14/2021, 1:21 AMWhat it does - t
Toasty
04/14/2021, 1:21 AMit automatically makes a dataset - u
user
04/14/2021, 1:21 AMFor tacotron? - t
Toasty
04/14/2021, 1:21 AMYes - t
Toasty
04/14/2021, 1:22 AMARPABET, transcripts, wavs, you name it - t
Toasty
04/14/2021, 1:22 AMOne problem - t
Toasty
04/14/2021, 1:22 AMSo, you're suppose to dump like several 22050 hz mono audio files of your speaker with a sample wav of said speaker. - u
user
04/14/2021, 1:23 AMDiscombobulate the allies of the genetical chromosomian ectomycorrhizals. - t
Toasty
04/14/2021, 1:23 AMASSFAP would take the sample wav and find all possible voice thingies that match the 10 second sample wav. - t
Toasty
04/14/2021, 1:23 AMThe problem? It has a godawful time at extracting the audio to wav.