Empty language sequence in the output

Hi, I am trying to test the scenescript model on my data but facing issues!
I am able to get the desired output from the ase and aea sample datasets as shown in the demo notebook.
But when I test the models on my own data which are also aria glasses recordings and MPS semidense pointclouds, I get empty language sequences but no errors in the code, wondering what is wrong in my recordings, the code does the preprocessing at inference so my data should match the trained datasets.
Thanks!

Start with the demo notebook and compare its ase and aea inputs with the author's Aria glasses recordings and MPS semidense pointclouds. Trace the inference-time preprocessing and identify why the custom recordings produce empty language sequences without errors; done means the cause is documented or the same inputs yield non-empty output.

Written by the indexing model from the issue text.