depends on the GPU

what GPU do you use?

Tesla V100-SXM2-16GB

i'd give it 1.5h

and then listen to the results

ok

2h is a decent amount of data so even training longer will be beneficial

I see.

I'm trying to make sure my validation loss low and this... slope thing looks good

Currently at 14 minutes.

yes it looks good for that amount of time

ye.

I’m hoping the update on the Dio model is gonna be pretty good. I added at least 7 minutes worth of dialogue. The original model thats on the site right now has only 85 wav files with more than 3 minutes worth of dialogue. Meanwhile the updated transcript I have shows to have 213. Once I actually splice up all the audio clips and train the model it should be good to go along with a few other models, some I stated before, some not.

What would be a reasonable timeframe for the model to train once I finish the voice clips?

My training is sounding a bit crispy. Should I continue (epoch 180) or should I take a step back and fix something?

Source:

Generated:

Does anyone know how to fix this error?

you have to put the audios in 16 bit PCM

I believe I did that already while I was making the new clips

Im not sure if I let one slip by or what but if thats the case that isnt gonna be fun

Mhm

Okay I checked and I did export them all in 16 bit PCM wav at 22050 Hz

are you sure it's mono audio

not stereo?

please send me your dataset so I can have a look at the format your files are in

Sure thing one moment

Oh wait

I see the problem now

I didnt pay attention to the side of the actual file, which says stereo and 44100 hz despite the top and project rate saying otherwise