search

wit-ai / wit

Natural Language Interface for apps and devices
https://wit.ai/
942 stars 91 forks source link

Final transciption results from dictation are too fast and splits a phrase in pieces. #2672

Open JuanObiJuan opened 1 year ago

JuanObiJuan commented 1 year ago

Hello. Im using The AppDictationExperience script in Quest and I have the same problem in german, english and spanish.

Even talking at a normal speed, withou evidents gaps, the dictation returns as final transcription the phrase in pieces.

When I say for example: "I understand your situation" the dication returns two as two complete trasncriptions "I undertand." and "Your situation." separatly.

Is something Im missing?

Here is my AppDicationExperience config

image

MokhtarKhorshid commented 1 year ago

Is this only occurring on device or does it happen in editor as well?

JuanObiJuan commented 1 year ago

Hi. It happens on Editor (oculus link using the headset mic input as the preferred mic input in the system) and it happens also in the build (Unity 2021 and oculus quest 2)

JuanObiJuan commented 1 year ago

I realised that It looks like the first phrase is correct. But after the first final transcription, when I receive the notification of interaction finished (but the service is still on and sending data) the rest of the phrases are splitted.

If I condfigure the MinKeepAliaveTime and MinTranscriptionKeepAlive to the max then is not spliting the phrases anymore.

image

MokhtarKhorshid commented 1 year ago

When there is a pause (silence) in the speech, we recognize that as the end of an utterance. You have some control over this using the keep alive values.