handle long utts but not depending solely on increasing beam

[cassiotbatista]

MFA seems to recommend increase the beam value for long files (some issues from 2019 but some are more recent). EasyAlign used to apply a macro segmentation pre-stage before actually aligning (c.f. sect 2.1).

Might be hard on normalized text, tho -- no punctuation, capitalization, linebreaks or annotations of any sort.

[cassiotbatista]

Another option to keep an eye on: https://github.com/srinivr/kaldi-long-audio-alignment