1

Looking for a keyword spotting api, I found the most promising is OpenEars + Rejecto plugin. I need to recognize a pretty simple dictionary, such as Up, Down, Left, Right. Followed the steps to run the sample project here.-

http://www.politepix.com/rejecto/

But no matter what I speak, the app always recognize the same words in the same order. I'm sure I'm missing something obvious configuring the project, but can't figure out what. Anyone succeeded running the sample Rejecto project or can point me in the direction of a better sdk for keyword spotting?

ssantos
  • 16,001
  • 7
  • 50
  • 70

1 Answers1

2

OpenEars developer here. Please use the OpenEars quickstart tutorial and the OpenEars documentation to learn how to use OpenEars for speech recognition before integrating Rejecto: http://www.politepix.com/openears/tutorial There is also more documentation at http://www.politepix.com/openears and within the downloaded OpenEars distribution.

The documentation on the Rejecto page you linked to above is for integrating Rejecto into an already-working OpenEars-enabled app, but doesn't attempt to cover how to do speech recognition with OpenEars and doesn't contain a sample app.

Update: the confusion was mostly due to my shipping the last version of the sample app with a confusing setting which I just fixed; sorry about that ssantos.

Halle
  • 3,584
  • 1
  • 37
  • 53
  • Thanks @Halle. I actually followed "Offline speech recognition which ignores spoken words which aren't in the vocabulary" tutorial, "Using LanguageModelGenerator+Rejecto" section with no luck. Anyway, tried `OpenEars sample project`, and it always logs "GO FORWARD GO BACKWARD GO LEFT GO RIGHT", even if i keep silent. – ssantos Sep 29 '13 at 15:06
  • 1
    Hi ssantos, what do you mean by "with no luck"? An app made from the tutorial can't have ended up with the vocabulary "GO FORWARD GO BACKWARD GO LEFT GO RIGHT", since the tutorial doesn't use that vocabulary as its example vocabulary. – Halle Sep 29 '13 at 15:10
  • I mean it doesn't seem to recognize any word I say, and repeats always the same output log. The project I'm running is not from the tutorial, but from the `OpenEarsDistribution` downloaded from the OpenEars homepage. – ssantos Sep 29 '13 at 15:13
  • 1
    My question was what you meant here: " I actually followed "Offline speech recognition which ignores spoken words which aren't in the vocabulary" tutorial, "Using LanguageModelGenerator+Rejecto" section with no luck". What happened when you used the tutorial? – Halle Sep 29 '13 at 15:16
  • I meant I followed the steps described in "Using LanguageModelGenerator+Rejecto" to integrate Rejecto plugin in the openears sample project. I understood the other sections were different alternative approaches. Should I follow any of the previous steps? Am I wrong using the sample project included in the bundle and I'd rather start with an empty project from zero? – ssantos Sep 29 '13 at 15:22
  • Exactly right, the tutorial is for a brand-new app. If you apply the tutorial instructions to the sample app, things will get a little weird :) . I strongly recommend using the tutorial rather than the sample app because it shows you how to create your own vocabulary. But I also checked into the sample app due to your post here and noticed that I accidentally shipped it with a test audio file with PocketsphinxController running automated recognition on it, so I've fixed that and I appreciate your making me aware of it. The fixed version of the distribution will be up in about 20 minutes. – Halle Sep 29 '13 at 15:26
  • Thanks @Halle, and sorry for the misunderstandings, I'll give the tutorial a try from scratch then. – ssantos Sep 29 '13 at 15:36
  • Super, and please feel welcome to ask questions at the OpenEars forums (http://www.politepix.com/forums) since it's a better format for extended discussion. – Halle Sep 29 '13 at 15:37