i think 111-135 are short

I have had a bunch of short audio files and it hasn't given me errors for them, just made the graph lower resolution

I see we're getting some s t o n k s on Jeff Bennett Kowalski...

This man got 78 pinned on this task bar 😂

yeah XD

I got like 100

It looks like a ghost

Does the sampling rate taken by tacotron have to be downsampled from 48k?

Is it usually something like 22050?

For some reason one of the aligners i'm using does not accept a list of wav files due to their input signal length being too small to resample from. So I changed the sample rate from 48k -> 44.1k and it resampled fine but i'm a little worried that this will impact the model using this data

Especially since it's only some files like around 400 files being resampled. I don't know if I should remove these files or resample them to a smaller rate

we train our models on 22050hz if that's what you mean

the 4chan bronies have notebooks for 48000hz but those are less stable with low data iirc?

Fr tho how come bronies are always the ones to have hq tts

oh they did it years ago

we're using their old tech

well "their" other researchers make the tech, 4chan bronies are just the ones who work on making it easy to use

I guess their stuff just sounds better when actually working correctly?

this tbh https://discord.com/channels/768215836665446480/825017563196686338/876355127865405480

I hate that the initials for this are assfap

@BMWM3GTR did you remove all Noisy and Very Noisy files from the Pony dataset?

I see that you’re sampling at 44.1k also

@UserHow's the execution coming along?

It's okay I think

Where is it now?

Are alignments impacted by different sample rates via their timestamps/phonemes?

Do you have any links to the 4 chan bronies

I've been trying to get mellotron to work and it would be good to talk to someone who knows about it