Support forum for the recognizers

In this forum users are welcome to create new topics regarding their experience with the recognizers released under the AVATecH and AUVIS projects. The team of developers from:

. The Max Planck Institute for Psycholinguistics, Nijmegen
. Fraunhofer HHI Institute, Berlin
. Fraunhofer IAIS Institute, Sankt Augistin

will monitor this forum and will try to answer any question and solve any problem.

Before posting your question, please do the following:

. Read the Recognizers Manual:

. Browse the already available posts and find out if your issue hasn’t been addressed yet.

The AVATecH and AUVIS teams

Dear all,
sorry for the silly question, but I have difficulties in actually finding the recognizers extensions files for Elan. Do I have to ask for them or they are publicly available?
Thank you!

Hi Francesca,

a number of recognizer extension configuration files are not yet included in the standard shipping of ELAN, but you can download a ZIP with the whole collection here:

This page also links some documentation. The “some facts” tab there is interesting as well, although it only lists AVATecH workshops. The workshops of AUVIS are less end-user oriented yet.

A similar, more low-level avatech page on offers downloads of recognizer configuration files in various formats. That second page also links the CLAM webservice websites where you can run recognizers in your web browser, mostly for project-internal use.

Hi Eric,

Can you tell me if these ‘recognizers’ will allow ELAN to recognize the audio component of a Windows *.mov file. The file when played had clear audio but when uploaded to ELAN does not play or recognize any audio. This is causing me considerable difficulty. I need to be able to upload this material right away to work with it.
thank you

Hi Lisa, ELAN plays audio from separate WAV files because those make it easier to visualize the amplitude curves. You may want to ask in the ELAN forum whether it is possible to bypass this setting. In general, exporting the audio to a WAV file is probably the recommended way.

Regarding the recognizers, they often use ffmpeg libraries internally: That makes them able to support multiple file formats. When a recognizer asks for audio input, ELAN will not consider your MOV file as a candidate.

You could try to cheat by editing the recognizer configuration file (for example YOUR_ELAN_DIRECTORY/extensions/YOUR_RECOGNIZER_NAME/recognizer.cmdi) from input type=“audio” mimetypes=“audio/x-wav” into input type=“video” mimetypes=“video/*” or mimetypes=“video/quicktime”. You can also specify multiple mimetypes separated by spaces. The audio recognizer MIGHT be able to process your file even if it has the wrong type. Do not change other properties. Note that you have to restart ELAN and that ELAN will now say that your recognizer is a video recognizer instead of an audio recognizer.

Please keep us updated about your experiments.

Regards, Eric

Good news for all recognizer users: ELAN 4.7.x already includes the recognizer configuration, so you no longer have to download a separate ZIP from the website!

Unfortunately I made a typo in the “Tag Vowels” configuration for Windows: The URL mentioned in recognizer.cmdi has to end in “avatech_mpi_tag_vowels”, not in “avatech_mpi_tag_Vowels”. You can correct this yourself, or download the updated ZIP, or wait for the next version of ELAN. This recognizer invokes Praat and does some post-processing for you, but Praat itself deserves all compliments for how good the recognizer tracks voice pitch. If you do not use that particular recognizer in Windows, you do not have to do anything anyway.

In related news, the server now runs an updated version of the video Hand Head Tracking / Human motion analysis recognizer. You do not have to change anything in ELAN to use it. If annotation results are not as good as expected, define an output video file and watch it, it will clarify in which way the recognizer “misunderstood” your video. Often, adjusting the skin color, background or rest position parameters is worth trying.

Please do not hesitate to give us feedback when you use the recognizers.