Speed up fastpitch preprocess

[bzp83]

I'm following this notebook https://github.com/NVIDIA/NeMo/blob/main/tutorials/tts/FastPitch_MixerTTS_Training.ipynb

and the preprocessing (pre_calculate_supplementary_data) takes a very long time on my dataset (10+ hours)...

Is there any way to parallelize it? It seems to process the files one by one... I tried changing num_workers from 1 to 16 but there's no difference...

thanks!