Bug: text_to_speech.ipynb final cell - possible infinite loop?

[Veronika271]

Description

I have been running all of the cells in the text_to_speech.ipynb tutorial sequentially, and the final one has been running for over 25 minutes now without stopping.

Steps to Reproduce

1. Open up text_to_speech.ipynb.
2. Run every cell in order.

Expected Results

I expected the final cell to run in 2-3 minutes.

Actual Results

It has been running for over 25 minutes now, and it looks like this is due to a very long loop which might be infinite. I have tried stopping it and restarting it and this did not help.
Here is a screenshot of what I see. The gray text on the bottom looks kind of repetitive, which is why I feel like there might be a long loop involved.

Additional Notes

It's possible I am not patient enough, so I will leave this cell running for a few hours and update or remove the issue if it ends up finishing!

[Veronika271]

Update: it took 53 minutes and 50 seconds, but it worked!

[satra]

@Veronika271 - did you use the gpu for this? it's a setting in google colab (runtime).

[Veronika271]

I didn't, but I am trying to rerun the tutorial on GPU right now to see what happens!

[Veronika271]

It took 2 minutes to run on GPU, but the voice was less accurate.

[satra]

interesting (which means precision may matter and it may be good to set it explicitly). thanks for the update. @fabiocat93 - just a ping in case this rings a bell.

[fabiocat93]

interesting (which means precision may matter and it may be good to set it explicitly). thanks for the update. @fabiocat93 - just a ping in case this rings a bell.

On the CPU, the model may use FP32 (higher precision but slower), while on the GPU, it might use FP16 (lower precision but faster). This difference could be why the TTS results were lower quality on the GPU