[Resolved] How to reject out-of-vocabulary utterances? – Politepix /forums/topic/language-recognition-problem/feed/ Tue, 23 Apr 2024 14:39:19 +0000 https://bbpress.org/?v=2.6.9 en-US /forums/topic/language-recognition-problem/#post-8198 <![CDATA[[Resolved] How to reject out-of-vocabulary utterances?]]> /forums/topic/language-recognition-problem/#post-8198 Thu, 01 Dec 2011 09:59:37 +0000 jerry e.g in the “.languagemodel”there are [eat, fat, love], when I said [ear], it will return [eat]. but I want to get a null result.
what should I do？thanks!

]]> /forums/topic/language-recognition-problem/#post-8199 <![CDATA[Reply To: language recognition problem]]> /forums/topic/language-recognition-problem/#post-8199 Thu, 01 Dec 2011 12:00:42 +0000 Halle Winkler Hi Jerry,

This is the OOV utterance problem — here is the relevant Pocketsphinx FAQ entry:

http://cmusphinx.sourceforge.net/wiki/faq#qcan_pocketsphinx_reject_out-of-grammar_words_and_noises

]]> /forums/topic/language-recognition-problem/#post-8203 <![CDATA[Reply To: How to reject out-of-vocabulary utterances?]]> /forums/topic/language-recognition-problem/#post-8203 Thu, 01 Dec 2011 12:33:56 +0000 jerry Thank you!
I will see that. ^_^

hi!

I saw the doc .But It’s difficult to understood for this: “However, since a phone loop should give us better rejection than confidence scores, I propose that we implement a phone loop in the FlatLinguist as a first attempt. It should be fairly easy to expand the unknown word <unk> into an all-phone self-looping network. As a starter, context-independent phones will be used, since context-dependency was shown to not matter in this case. If a result contains the <unk> word, the utterance can possibly be out-of-grammar. We leave it to the application developer to determine that. “.

1.what the “phone loop” mean?

2.how to use “<unk>“? Is there a sampleCode for Implementation Details?

thx a lot~

]]> /forums/topic/language-recognition-problem/#post-8209 <![CDATA[Reply To: How to reject out-of-vocabulary utterances?]]> /forums/topic/language-recognition-problem/#post-8209 Fri, 02 Dec 2011 06:50:52 +0000 jerry hi Halle,
I saw the doc .But It’s difficult to understood for this: “However, since a phone loop should give us better rejection than confidence scores, I propose that we implement a phone loop in the FlatLinguist as a first attempt. It should be fairly easy to expand the unknown word <unk> into an all-phone self-looping network. As a starter, context-independent phones will be used, since context-dependency was shown to not matter in this case. If a result contains the <unk> word, the utterance can possibly be out-of-grammar. We leave it to the application developer to determine that. “.
1.what the “phone loop” mean?
2.how to use “<unk>“? Is there a sampleCode for Implementation Details?
thx a lot~

]]> /forums/topic/language-recognition-problem/#post-8211 <![CDATA[Reply To: How to reject out-of-vocabulary utterances?]]> /forums/topic/language-recognition-problem/#post-8211 Fri, 02 Dec 2011 08:31:02 +0000 Halle Winkler Hi Jerry,

Since these are questions about the implementation of Pocketsphinx it would be better to ask them at the CMU Sphinx site.

]]> /forums/topic/language-recognition-problem/#post-11115 <![CDATA[Reply To: How to reject out-of-vocabulary utterances?]]> /forums/topic/language-recognition-problem/#post-11115 Fri, 14 Sep 2012 15:03:09 +0000 Halle Winkler Updated: you can now handle out of vocabulary rejection using Politepix’s Rejecto plugin for OpenEars.

]]>