[Resolved] Rejecto delivers rejected speech

[noam]

Hi,
I am running Rejecto using a prebuilt dictionary.
But [pocketsphinxDidReceiveHypothesis:recognitionScore:utteranceID:] keeps calling for rejected speech.
Log for example:
The received hypothesis is ___REJ_A ___REJ_A with a score of 0 and an ID of 1

Any known reason for that?

Thanks,
Noam

[Halle Winkler]

Hi Noam,

Rejecto doesn’t operate with prebuilt language models, it is for creating a language model at runtime while running Rejecto.

[noam]

Thanks! That fixed it.

Is there any way to reduce the sensitivity of OEPocketsphinxController?
Version 2.0 is much more sensitive then 1.65

[Halle Winkler]

Super, glad that helped. You can increase or decrease OEPocketsphinxController’s property vadThreshold. Higher settings will exclude more noise (but if it’s too high, it will exclude non-noise speech) and lower settings will allow more noise. I am still searching for the correct default setting for vadThreshold since it is part of the brand-new area of the framework, so your feedback on that is useful. OpenEars’ current default value is 1.5 and CMU Sphinx uses the default value 2.0 (but I thought it was excluding a bit too much real speech).

[noam]

Thanks a lot for all your help!
Great framework!

[Author]

Posts