# machine-learning
- u
[K$ng] ZombieGeneral
06/18/2021, 7:34 PMLET GOOO - m
mega b
06/18/2021, 7:39 PMnice - u
[K$ng] ZombieGeneral
06/18/2021, 7:40 PM@User this is auc using autogenerated subtitles with a dataset builder - u
[K$ng] ZombieGeneral
06/18/2021, 7:40 PMLol - m
mega b
06/18/2021, 7:40 PMoh wow! - m
mega b
06/18/2021, 7:40 PMi think jacksepticeye's model was also made automatically - m
mega b
06/18/2021, 7:40 PMwithout tweeking the dataset - m
mega b
06/18/2021, 7:40 PMso it should sound decent - u
[K$ng] ZombieGeneral
06/18/2021, 7:59 PMdecent is an understatement - u
[K$ng] ZombieGeneral
06/18/2021, 7:59 PMholy shit it doesnt even sound like an ai - m
mega b
06/18/2021, 7:59 PMright well the process of dataset making can be sped up to handle more voices - u
[K$ng] ZombieGeneral
06/18/2021, 8:00 PMhm seems like it decided to fuck itself over - u
[K$ng] ZombieGeneral
06/18/2021, 8:00 PMprob gonna get better tho - p
pi
06/18/2021, 8:03 PMit's just the alignment graph - p
pi
06/18/2021, 8:04 PM😐 - u
[K$ng] ZombieGeneral
06/18/2021, 8:04 PMye but its still a good sign - c
CupOfGeo
06/18/2021, 8:04 PMwow is this your voice? - u
[K$ng] ZombieGeneral
06/18/2021, 8:04 PMnope its on the site tho - u
[K$ng] ZombieGeneral
06/18/2021, 8:04 PMI'm doing DarkViperAU's voice - p
pi
06/18/2021, 8:05 PMit's not the old testament god - u
[K$ng] ZombieGeneral
06/18/2021, 8:05 PMwell I guess I did speak too soon that itll sound great - u
[K$ng] ZombieGeneral
06/18/2021, 8:05 PMthe graph flatlined on me - u
[K$ng] ZombieGeneral
06/18/2021, 8:06 PMFinally - u
[K$ng] ZombieGeneral
06/18/2021, 8:06 PMit improved - c
CupOfGeo
06/18/2021, 8:09 PMhow does your dataset builder work. does it transcribe and then segment it into clips? - u
[K$ng] ZombieGeneral
06/18/2021, 8:11 PM - u
[K$ng] ZombieGeneral
06/18/2021, 8:11 PMthats what im using - c
CupOfGeo
06/18/2021, 8:24 PMwow this is really cool if i could force align text and audio then that would solve a huge problem then if you could just find a model that can distinguish speakers - p
pi
06/18/2021, 9:23 PMthe problem with this as I see it is that TV expressive TV voices are very far out of distribution typically