# tacotron-2-support
- r
Reclezon
10/21/2022, 10:56 PMSelect any of them, or the latest one to test. Later ones (assuming you haven't been training far too long) will sound better - e
Eugene
10/21/2022, 10:57 PMI'm in that Synth notebook/Colab now, but I'm kinda confused with all the fields and stuff. - e
Eugene
10/21/2022, 10:58 PMI'm not sure if I should leave this model_id value to what it says now or that I should, somehow, import my model there. - u
{K EY1} (Kei)
10/21/2022, 10:58 PMShare the pt file on google drive Then paste the link and cut out everything but the middle part with numbers and letters - r
Reclezon
10/21/2022, 10:59 PMThqta how you import your model. Only the drive Id (ex: 1--LZFzYiU-uxkR309PSim_RWWLA2-bqAnecce) is necessary - e
Eugene
10/21/2022, 10:59 PMAight - e
Eugene
10/21/2022, 11:00 PMAnd then I only have to execute this after that? - e
Eugene
10/21/2022, 11:00 PMAnd it will save an audio output file somewhere on my Drive? - r
Reclezon
10/21/2022, 11:00 PMItll display it inline in the log output - r
Reclezon
10/21/2022, 11:00 PMBut yes - r
Reclezon
10/21/2022, 11:01 PMSpeaker weigting isn't necessary unless you're training multiple voices at once - e
Eugene
10/21/2022, 11:01 PMAight - e
Eugene
10/21/2022, 11:01 PMAlso the validation loss is kinda swinging between 0.12 and 0.13 now. - e
Eugene
10/21/2022, 11:02 PMI haven't seen that value go up again before, till now. - e
Eugene
10/21/2022, 11:04 PMOh, and I also just noticed a warning 😅 - e
Eugene
10/21/2022, 11:05 PMThat's odd, I actually made sure beforehand that the audios didn't override the 0db limit in Audacity. 😅 - e
Eugene
10/21/2022, 11:06 PMAh, it's finally continuing with 0.11 now. - r
Reclezon
10/21/2022, 11:06 PMThat's normaal. Idk why. Probably output audio exceeding volume limits - e
Eugene
10/21/2022, 11:07 PMHmm - e
Eugene
10/21/2022, 11:07 PMAlso, I got a new error here - e
Eugene
10/21/2022, 11:07 PMIn the Synth Colab. - r
Reclezon
10/21/2022, 11:08 PMThe
line expects the model to be multispeaker. For just one voice, jsut clear the line.speaker_weighting - e
Eugene
10/21/2022, 11:10 PMIf the zoo bans me for hollering at the animals, I will face God and walk backwards into hell! - e
Eugene
10/21/2022, 11:10 PMWelp here's the result at val loss 0.12 and 100 epochs. - e
Eugene
10/21/2022, 11:11 PMIt works, but without the text written down under it it's a little hard to understand. - e
Eugene
10/21/2022, 11:13 PMAlso, the voice sounds a bit.. well, actually not that robotic, but it does sound kinda low-quality. But it only has 94 wav files to work with of course. - u
{K EY1} (Kei)
10/21/2022, 11:14 PMPut
in the box0:1 - u
{K EY1} (Kei)
10/21/2022, 11:15 PMOh you already got it to work oops - e
Eugene
10/21/2022, 11:15 PMThe audio quality sounds much better now! But the intonation sounded more natural in the first one. - u
{K EY1} (Kei)
10/21/2022, 11:15 PMSometimes it has different intonation each time you generate it