Determining Word Similarity

[stressdesign]

Hi,
I understand that while the application is able to determine the best sound match(s) to words in the grammar it would be helpful to understand how closely words in our grammar are inherently. E.g. “Jim” sounds somewhat similar to “Tim”, whereas “Jim” sounds nothing like “Mike”.

Clearly, the application interprets spoken phonemes and creates a scored match to words in the grammar. Is this data easy accessible in the files generated?

If we can easily determine likeness up front, it can enable us and other developers to programmatically find non-unique words and force disambiguation.

[Halle Winkler]

Welcome,

Are you asking for scoring?

[stressdesign]

Yes.. specifically a quantified value of phonetic closeness in words.

I know the app scores inputed audio to words in the grammar, but a score or value of matches between words in the grammar would help to establish a more unique grammar.

[Halle Winkler]

Scoring is returned with most APIs, although its applicability is inherently limited. You can search these forums for more info about that.

a quantified value of phonetic closeness in words

This is a score.

To clarify, when you refer to the app or application, are you referring to someone’s particular app implementation of the OpenEars framework?

[stressdesign]

To be clear, i’m not asking about the score between a spoken word and the matching words in the grammar. I’m able to see that already. What I’m asking about the relative phonetic closeness of words in the grammar.

[Halle Winkler]

Their closeness to each other?

[stressdesign]

exactly.. asking the app to discern between tim and trim is much harder than between tim and time. tim and trim are phonetically similar. It would be very helpful to have that quantified easily to assist (possibly by warnings) of grammars that need altering.

[Halle Winkler]

Got it. This is getting a little bit beyond the scope of an ASR tool and more into the region of text analysis. As you dive into a goal like this, many questions start to come up about what is “good handling” for advanced cases and then it gets multiplied by the different requirements for language models versus grammars versus RuleORama grammars, and then things that may or may not be equally likely across languages. It is also something that can usually be judged by visual observation of your own grammar.

In an individual app (versus a framework like OpenEars) you can probably restrict the range of what is likely much more, making this simpler to implement for your own specific case. You can check out the file LanguageModelGeneratorLookupList.text in the acoustic model for the language you’re using and (for instance) load it into a data structure like an NSDictionary in order to be able to access it to do evaluation of closeness according to the needs of your application, if there is no opportunity to just look at the grammar at the time of creation and consider whether it contains similar words.

[stressdesign]

ok, i’ll give that a shot. thank you!

[Halle Winkler]

You’re welcome!

[Author]

Posts