for hifigan 5k steps (1250 epochs) for talknet+hifigan 2k steps

almost there

sometimes looks like this tho

can someone test this synthis and tell me if it works https://colab.research.google.com/drive/1vAo9eClAnpW2m5ocekHw1PhC8VvvGk7K?usp=sharing

Should I use a model while it's still training?

no

if its not training and saving too fast you can use it tho

i get this every time i synth

what on earth is this voice

sounds like a chihuahua

.................................................................................................................................................................................

sounds like he's in pain

um

robot

lol

not sure why it sounds so robotic

that's what I'm doing tbh, it's a good way to gauge how the training is going and how much longer it still needs to run

ay i got a transcript that transcribes the files like "wavs/audiofile|audiofilesize|text", any easy way to make it just "wavs/audiofile|text"?

its 3625 files no way im doing this manually

awk -F '|' '{print $1"|"$3}' old_file.txt > new_file.txt

i dont think i'll ever make these models public due to a few reasons, but i made michael and cary huang talknet models

michael's model has the issue of being overly quiet (i had to amplify the audio on the sample I sent here) because the model was mainly trained on a livestream that was kinda quiet

Hi guys! At 15.ai you have a list of most confident audios. Does smb know how many you should pre-generate and pick the best? I see top3 at the site. Also is there algorithm to increase confidence? I see some hacks with posterior distribution of alignments, where can i read about that?

almost: --------------------------------------------------------------------------- FileNotFoundError Traceback (most recent call last) in () 41 42 # Load WaveGlow ---> 43 waveglow = torch.load(waveglow_pretrained_model)['model'] 44 waveglow.cuda().eval().half() 45 for k in waveglow.convinv: 2 frames /usr/local/lib/python3.7/dist-packages/torch/serialization.py in __init__(self, name, mode) 209 class _open_file(_opener): 210 def __init__(self, name, mode): --> 211 super(_open_file, self).__init__(open(name, mode)) 212 213 def __exit__(self, *args): FileNotFoundError: [Errno 2] No such file or directory: '/content/waveglow.pt'

training hotel mario luigi

I had to use some extra sources

not sure if it's enough though

then again