Using openwakeword tflite with flutter

[hasanalaniIT]

Hello guys I'm building my own virtual assistant and I'm facing a problem with using the wake word due to the shape of the model

what I'm trying to do is
to use a microphone that passes raw data byte stream every second on a sample rate of 16000 HZ and length of 16000
but the shape of the model is (1, 28, 96) I tried to reshape my audio data but it does not work because it needs audio data length or size of (2688)
so this is the simple flow that I want to achieve :
mic data 16000 len --> openwakeword tflite model -->> result --> repeat on no detection
what are the steps needed to achieve this flow or behavior with openwakeword models?

[dscripka]

Using openWakeWord actually involves 3 separate models working together: a melspectrogram model, an embedding model, and then the wakeword prediction model. You can see how all three of these combine together here.

Re-implementing this in flutter is possible, but you'll have to match the logic that prepares the audio and passes everything through these three models in the proper order.