Time, place: 9:30-11:15, CSC

Present: Juho, Mietta, Martin (memo)

Agenda

aalto-asr

What to do with it, how to support it?

Status of André’s old scripts?

Decision: We phase out aalto-asr and declare it no longer supported.

kaldi+kaldi-asr

Kaldi installed on Mylly2, kaldi-asr status unclear. Does it work for alignment?

Decision: We provide a script that for example:

kaldi-asr –align –language fi –number-of-speakers N –audio-file input.wav –textfile input.txt -output-format txt|eaf|srt –out aligned text

Needed

• Ansible script to install kaldi-asr
  • first install kaldi
  • copy/link bin steps utils
  • models from Juho in zip file
  • kaldi-asr needs work, maybe implement as Ansible template
  • First set of languages: fin,swe,eng (later: saami, komi, etc)
  • first step: Check if present kaldi-asr works as recognizer after install, add –align later.

Other issues?

Ideas for later: Users can create derived models for new languages, if they provide phoneme-grapheme mappings.