# machine-learning
- b
Balloonya from Pokémon
05/26/2021, 11:23 AMI have synthesized my model of a voice of the announcer but it gives me a bad results. It was because of poor quality. I had 78 audio files and I don't know what to do to reduce the background noise. - u
user
05/26/2021, 11:23 AMim gonna chalk it up to my data sucking - b
Balloonya from Pokémon
05/26/2021, 11:33 AMNever working. - p
pi
05/26/2021, 11:45 AMand now explain in detail - b
Balloonya from Pokémon
05/26/2021, 11:49 AMOkay. I have created my own model of the Tacotron 2, but then when synthesizing it seems like I don't understand. In other words, it makes a lot of noise to not understand each other. It gives me problems with 78 audio files, that in the background there is a bit of noise. - b
Balloonya from Pokémon
05/26/2021, 11:49 AMIt was happened in Google Colab. - b
Balloonya from Pokémon
05/26/2021, 11:50 AMThe Validation loss of my model is above of 0.1 - b
Balloonya from Pokémon
05/26/2021, 11:52 AMI have 5 Epochs in the Tacotron 2 model on the Google Colab notebook. - p
pi
05/26/2021, 11:58 AM5 epochs isn't enough - b
Balloonya from Pokémon
05/26/2021, 11:58 AMAh, that's right. - b
Balloonya from Pokémon
05/26/2021, 12:25 PMStill not working. Too hard for me. - a
Alexius08
05/26/2021, 12:40 PMInitial training with the whole dataset for Gandhi (12 minutes transcribed, 12 minutes pending to be transcribed) went badly. For this session, I used only a 6-minute portion of the dataset, and it's going well. - s
SidPlays_144p
05/26/2021, 12:40 PM👀 - s
SidPlays_144p
05/26/2021, 12:40 PMthat's the slopiest slope i've seen today - l
LindosoLord
05/26/2021, 12:48 PMThis is going nicely - l
LindosoLord
05/26/2021, 12:48 PMI guess - u
user
05/26/2021, 12:58 PMstarting out nicely - u
user
05/26/2021, 12:58 PMmaybe this time it wont be a bunch of mumbling - a
Alexius08
05/26/2021, 12:58 PMAnyway, here's how badly my earlier attempt for training a Gandhi model went. Has anyone here encountered an alignment chart like this before? - u
user
05/26/2021, 12:58 PMwoah - u
user
05/26/2021, 12:59 PMwhats a good validation loss number when its done training on average? - a
Awesome Face (GTX 1660 Ti)
05/26/2021, 12:59 PMsome where less than 0.1 - u
user
05/26/2021, 1:00 PMalright thanks - p
pi
05/26/2021, 1:00 PMNo - u
user
05/26/2021, 1:01 PMthis time around i'm not only using more data, but deleted some of the garbage - l
LindosoLord
05/26/2021, 1:01 PMDon't tell me it went like shit - p
pi
05/26/2021, 1:02 PMWell epoch 11 - a
Alexius08
05/26/2021, 1:03 PMAfterwards, the validation loss decreased but the chart is almost completely black. - a
Alexius08
05/26/2021, 1:03 PMThe line stopped midway. It's supposed to go all the way to the opposite corner. - l
LindosoLord
05/26/2021, 1:04 PM...