Can anyone please help me find out why my audio is soundlike like a robotic bumblebee with tinnitus in the background? I don't know what I did wrong here

YIKES

incorrect sample rate

make sure all your clips are 22050hz

Hey guys , is it possible to smooth out the robotic sounds in a generated voice to make it sound better without making a new voice training from scratch, even though the overall voice sounds accurate and similar to the original ? If yes then how to do it ? Thanks

maybe using voicefixer, id rather train a hifigan model just for that voice

not sure what dawn is onto, i already trained multiple models with 22050hz (talknet converts them already for you and bing always forgets to convert them manually) pretty sure your problem is because of the problem you mentioned before on here, hifigan started training on an extremely high step, you shouldve deleted the hifigan folder from your drive, made sure that colab refreshed and knew that the folder was deleted, restart the runtime run step 3 and tried the hifigan training step again

Pixray presents: "A rubber duck with a Santa hat on its head"

what im getting at is i forgot to change some wavs in my dataset to 22050hz and i got the same issue

Ah okay thanks! Yes I double checked and they are all 22050hz. I did delete the hifigan folder but I did not restart the runtime so that’s what probably happened. Also I used only 40 wav files to test but would training those files to 6000 steps be too much? Should I just do lesser steps?

i dont think you can go over 3100 steps without some modification to the code, and since its only 40 wav files it wont improve much training more than 2000 steps

6000 steps is actualyl about what id shoot for

thats only with criminally low wav numbers

when the wav count is too low it doesnt even do steps its annoying tbh 😔

So if one day I have the time to make like 500+ audio files would 6000 be what I should train to ?

wait what the fuck? its hard locked at 3100 now like completely?

i thought that was just cause i had 4 wavs what the fuck

why

idk if you want so badly for hifigan to actually train id dupe the audio files (its not new data at all but if you want hifigan to atleast have an idea of your voice then)

if you have a lot of audio files and you want to guarantee a badass model then hell yeah

I see an Asian woman sitting and a bottle on the table

And I see books📚

Getting ready for SpongeBob SquarePants in Polish Dataset

just posting this here for general lol

The voice model will be in the "SpongeBob Kanciastoporty" category. 🙂

is there any deepfake image colabs that aren't obtuse?

RubberGoose

please I just want to swap vinny with the monkey from jimmy neutron

what the fuck