But I think the audio quality is somewhat close

Agghh dang

I think theres amd optimizers too

But cuda is really top of the line

But ye give me feedback on the SV2TTS model I trained, curious how it compares to tacotron2. Its using tacotron and WaveRNN

?

Is the Rise and Shine Mr Freeman one SV2TTS?

Yep

That one does actually sound pretty good

How long did training take?

Uuu

Ah hmm could take a look

On my 1660 with half the batch size it should've been, 1.5 days on the synthesizer, and .5 days in the vocoder

Was gonna buy a good setup for like $2k but my stonks didnt mature.

Training data was 0.9hrs

Am I going insane or what's with the middle finger reacts

Ok, then real time inference?

Someone might be mad we're sending lots of messages? lmao

??

Dont be passive aggressive dude just let me know if theres an issue

@User ?

Oh wtf. Yea mods need to ban them.

Wew

Inference time is constant per string length, counts as real time

I've seen faster inference tho, GlowTTS is faster

ah ok cool

Yea it can be difficult to remember how each sound to compare and balance with their perks.

Yeaaaah

I dunno if I should switch architectures

I notice the SV2TTS doesnt have as many 'strokes' like the other models who struggle with stop token prediction