Akash Singh

Results 26 comments of Akash Singh

@niartnelis did you figure out the data preprocessing part? and dataset format?

@JoeyHeisenberg how did you set up for multi-GPU on a single system?

@Jeevesh8 i tried it but gets stuck for hours after Done initializing distributed.

@Ahmad-noborders were you able to resolve the issue?

@rafaelvalle I have a similar setup for Hindi, I have trained for about 230k steps but the attention is not aligning. I have a trained tacotron model for Hindi which...

![losses_500k](https://user-images.githubusercontent.com/18519731/88510374-7c2b2080-d000-11ea-95e7-9b1905a52a8c.png) ![attention_500k](https://user-images.githubusercontent.com/18519731/88510375-7d5c4d80-d000-11ea-9ebb-d7d8c30e3a1a.png) My loss curves and attention looks like this post-training for 500k steps.should I decrease the learning rate? Any suggestion @akshay4malik @rafaelvalle as audio generated by model is gibberish

Hi ibro45, Thanks for your help, I also thought that this might be the case as I was providing 2-3 seconds of the audio file to predict. I will make...

Yes, I was thinking the same of Separating Encodec as a different module as it could be used individually and in many TTS systems like VALLE and VITs.

@awni I just started yesterday night and there are modules which are directly available in torch like LSTM and sequential layers which are used in ENcodec but not available in...

Yes i checked that PR today morning. it has most of the things. i will go through Encodec code and come back with more details. Maybe by tomorrow. @awni just...