Reply To: Need for word and phrase recognition with additional online processing

Home Forums OpenEars Need for word and phrase recognition with additional online processing Reply To: Need for word and phrase recognition with additional online processing

Halle Winkler


Sure, I can give a try to answer your questions.

1) Will open-ears run off an ipad be able to effectively recognise my keywords using south african english and to what accuracy level do you expect that it can deliver?

For keywords, I would expect that the answer would be yes. It isn’t possible to predict accuracy levels because they are very app- and vocabulary- and user-specific. I recommend doing a bit of testing, reading some of the discussions here about refining accuracy using the tools available in the SDK (such as choosing a language model or grammar, picking words which don’t have many rhymes, taking care with the vadThreshold, maybe using Rejecto) and if you aren’t happy with the accuracy after that, the last step can be to do some adaptation of the acoustic model with the accent in question, but this really may not be needed with a small vocabulary.

2) I would need to be able to assign key words and phrases to different messages stored within my app that the speaker will be speaking around. Could this be handled by an implementation? so if keywords / phrases were uploaded online and delivered to the app and assigned to a message could this work effectively?

I don’t quite understand this question yet, can you elaborate/clarify?

3) I would like to be able to store and upload the wave file at the end for full online conversion and future reference. would this be feasible and what server based engine do you suggest for most accuracy

I think you should be able to do the saving part with SaveThatWave. All of the cloud speech services in our ecosystem are equally esteemed by me and I could hardly bring myself to choose between them.

4) I would need to stream the audio live to the emotion analytics engine. would this be feasible?

How live? Live like “the moment the user is done speaking” or live like “as they are still speaking”?

5) Do you know of any open ears experts that would be able to assist me in delivering elements of this demo ASAP?

I don’t keep a list, but you are the fourth app producer who has asked me this in the last week so it’s clearly time for me to create a jobs board or some form of recommendation service. I have to give some thought to how to do this so it’s of maximum utility to all parties.