How to use timesteps information to calculate word-level alignments?

[abarcovschi]

I am trying to create word-level time alignments from the output of the beam search decoder, but I am not sure how to use the timesteps field to calculate this information. Could anyone please provide some formula or code implementation on how to convert the timesteps into word-level time alignments? I would be extremely grateful for any pointers!

[farisalasmary]

Hi @abarcovschi,

You can use this library to get the word boundaries
https://github.com/lumaku/ctc-segmentation

[abarcovschi]

Thank you!