Uberduck

I've never differentiated with numbers before
I'm mostly familiar with arpabet in the context of singing synthesis
Do the numbers denote stress?

0 for none, 1 for primary, 2 for secondary

I can see why oy0 doesn't have many matches then

I can't think of a single word that would use it

`ballpoint` is one of the examples i found

Huh
I wouldn't have thought that word has any unstressed syllables

it does according to cmudict ¯\_(ツ)_/¯

Do you stress the first syllable when saying it? I say it all evenly

<@!327116758714417153> what is the name of the app you found?

unfortunately it's only on android as far as i know

results are predetermanistic, by that I mean the synthesis sounds the same each run

Ok *that* is interesting
Must be why they're so small
It's got some stuff already calculated ig

nah it's probably something like fastspeech

idk why tacotron isn't deterministic actually

thats my nitpick about deepponies
my favorite aspect of it is that it really seems to respect exclamation marks

called it, it's based on fastspeech2 https://desuarchive.org/mlp/thread/38108843/#38197828 https://arxiv.org/pdf/2110.12612.pdf

```Specifically, for 48 kHz modeling, we predict 16 kHz mel-spectrogram in acoustic model,
and propose a vocoder called HiFiNet to directly generate 48 kHz waveform from predicted 16 kHz mel-spectrogram, which can better trade off
training efficiency, modelling stability and voice quality.```

so they train a 16khz mel spectrogram then use their HiFiNet to upsample basically