AuToBI error: No feature extractor registered for feature

[Gregory Durrett]

Hi,

My name is Greg; I'm a grad student at UC Berkeley working with Dan Klein. We're trying to use AuToBI to predict break indices for CTS data, currently Switchboard. We used the Games models to try to run on the attached files:

java -jar AuToBI.jar -input_file=text/sw2018.A.alignments.TextGrid \

-wav_file=audio/wav/sw02018.A.wav \

-pitch_accent_detector=models/games.acc.detection.model \

-pitch_accent_classifier=models/games.acc.classification.model \

-intonational_phrase_boundary_detector=models/games.inton_phrase.detection.model \

-intermediate_phrase_boundary_detector=models/games.inter_phrase.detection.model \

-phrase_accent_classifier=models/games.phrase_acc.classification.model \

-boundary_tone_classifier=models/games.pabt.classification.model \

-out_file=text/sw2018.A.breaks

The output is as follows:

1 [main] INFO edu.cuny.qc.speech.AuToBI.util.AuToBIUtils  - Reading words from: text/sw2018.A.alignments.TextGrid

edu.cuny.qc.speech.AuToBI.core.AuToBIException: No feature extractor registered for feature: nominal_PitchAccentType

at edu.cuny.qc.speech.AuToBI.AuToBI.initializeReferenceCounting(AuToBI.java:308)

at edu.cuny.qc.speech.AuToBI.AuToBI.extractFeatures(AuToBI.java:263)

at edu.cuny.qc.speech.AuToBI.AuToBI.run(AuToBI.java:1390)

at edu.cuny.qc.speech.AuToBI.AuToBI.main(AuToBI.java:1298)

We did some digging around and tried to manually register the PitchAccentTypeFeatureExtractor, but that just led to another error. It looked like a bunch of required feature extractors were not registered, including for features like max[subregionC[delta[rnormC[I]],psyl]] where it was not clear what should be added (the deprecated registerAllFeatureExtractors didn't work). We got the same error from using today's version of the code from GitHub.

What are we doing wrong? Any help would be appreciated!

Thanks,

Greg

[Andrew Rosenberg]

Ack, I'm so sorry Greg.   I'll be able to put some cycles on sussing this out tomorrow, and I'll get a fix up.  I'm sure it's something dumb.

Andrew

--
You received this message because you are subscribed to the Google Groups "AuToBI Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to autobi-users...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Andrew Rosenberg]

Good news: I found where the problem is.  It was just a boneheaded mistake on my end, and a somewhat hasty push of version 1.4.  Testing should have caught it, but didn't.  I apologize again for wasting your time because of my error.

Bad news: I'm on an airplane and am having a tough time retraining models and pushing code from the wifi.    I'll do my best to get version 1.4.1 uploaded as soon as possible -- monday at the latest.  Thanks for your patience.

[Greg Durrett]

No worries; thanks a lot for looking into this! Glad to hear that there's a fix, and we look forward to getting that.

Greg

You received this message because you are subscribed to a topic in the Google Groups "AuToBI Users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/autobi-users/oPIhDbY9KaE/unsubscribe.
To unsubscribe from this group and all its topics, send an email to autobi-users...@googlegroups.com.

[Greg Durrett]

Hi Andrew,

Sorry to pester, but did you get a chance to push that fix yet?

Thanks,

Greg

[Andrew Rosenberg]

No pester at all.

Ultimately, it's made more sense to roll the fix into a broader release.   The next version of AuToBI will be under the Apache 2.0 license rather than GPL.  This has been in the works for the last few months.  

I'm at interspeech this week, but should be able to get this pushed today or tomorrow at the latest. Models have been retrained and the code is ready.  I just have to put the finishing touches on a script to update the license notices on the source files.

Sorry for the delay.

-Andrew

[Greg Durrett]

Cool, thanks!

Greg

[Andrew Rosenberg]

Greg and all other AuToBI users.

Version 1.5 of AuToBI is currently available on github, and a compiled jar is available at http://eniac.cs.qc.cuny.edu/andrew/autobi/ along with models trained on the Columbia Games Corpus.  

The biggest change from version 1.4 to 1.5 is the change of license from GPL 3 to Apache 2.  This required a fairly substantial rewrite of some utilities that were covered by the GPL, but hopefully will make AuToBI more useful for more users.

I will be training new models over the next week or so and will make these available on the AuToBI website as they come online.  If there are any that you would like to see prioritized, please let me know.

-Andrew

[Jesin James]

Hi,

Did this error get resolved in the new version(1.5.1). I am trying to do the same, but facing the same error, still. Please comment

Regards

Jesin