IPS-LMU / EMU-webApp

The EMU-webApp is an online and offline web application for labeling, visualizing and correcting speech and derived speech data.
http://ips-lmu.github.io/EMU-webApp/
MIT License
51 stars 14 forks source link

anagest in webApp does not correctly detect gestural landmarks #322

Open samkirkham opened 2 years ago

samkirkham commented 2 years ago

note: I'm completely aware that this project isn’t actively supported, but posting this here for (i) documentation of the issue; (ii) in case somebody can point me in the right direction to fix it myself (which I am very happy to attempt, if anybody has an idea about what's going wrong).

In EMU webApp, selecting a portion of the tangential velocity signal and running anagest almost always returns ‘Could not find any values that step over the threshold!’, even when there is a sensible velocity trajectory in the selected interval (e.g. see the selected interval below). You can get gestural landmarks eventually if you mess with the selection enough, but they're not always correct.

Screenshot 2022-06-08 at 10 11 58

The anagest settings in my config file are all in line with the published examples, but I also get this problem on the example databases, such as ERC_SCATS_Polish at https://github.com/IPS-LMU/EMU-webApp/tree/master/src/testData/newFormat

See here for a local copy of the above dbs with minimal code: https://www.dropbox.com/sh/2ol0fvlzxxy8hme/AAALUajKzbWSns8OzuVMnzVla?dl=0

I was in email correspondence with @raphywink last summer and he confirmed that “unfortunately the current gestural landmark detection component seems to be broken entirely”.