Microphone audio feeding issue to the ONNX model

Hello all,
I am new to Unity and finding difficult to read audio input from mic to feed into the ONNX model in-order to predict the correct uttered alphabet.
So, I trained my audio model in Jupyter notebook, converted it to tflite and then converted it to ONNX model.

I enabled the microphone audio in Unity and could playback the voice recorded, and the next step is to get the uttered word displayed on the screen.

Please help me to fix this problem.
Any leads are highly appreciated.

Thank you,
Karthi B

 

Hi, since you already have a trained model for audio, you should try Barracuda for inference.
https://docs.unity3d.com/Packages/com.unity.barracuda@3.0/manual/GettingStarted.html
https://github.com/Unity-Technologies/barracuda-starter-kit
https://forum.unity.com/forums/barracuda.500/

ML-Agents doesn't support audio observations out of the box. You would need to implement a custom sensor for that, see https://github.com/mbaske/ml-audio-sensor