[In progress] Speed-up of marytts: New output_type with the information needed by wikispeech.js

Closed, DeclinedPublic1 Estimated Story Points

Actions

Some output_types in marytts add information after synthesis. Ex REALISED_DURATIONS. Use this to extract timing information and name of audio file. Use to avoid running synthesis twice, first when getting timing info and then when playing audio.

Working on this. TODO:
finish the code itself.
test.
comment here what was done.

Nearly there. There is now a marytts output_type WIKIPEDIA_JSON, that does the right thing.
Need to change the js to take
"tokens": [{"string":"XXX", "end":123}]
instead of
"tokens":[["XXX",123]]
New format is better!

Also there is a problem in marytts: With the tempfile written in HTSEngine output_type AUDIO no longer works. No message, everything looks right, but no audio is sent back to the browser.
Could have to do with things in Request. Hm. Fix this, at least with a workaround..

No longer relevant. Close.
But if we're still using marytts we still need to consider the speed issue later.