Use for audio models

[manena]

Hello, this is a great project! thanks!
I don't have a lot of experience on deep learning. I would like to use a tflite speech command recognition model I downloaded with your example project, but I'm having trouble with the input shape.

I've been trying different things and I sometimes get:

• That the input must be an array of primitives
• TensorFlowLite operation failed
• Sometimes it just crashes and unity closes.

Have you tried any audio model on your project? Is there a known issue with it?

Any clue that helps me fix it will be more than welcome!

[manena]

The problem seems to be when calling TfLiteTensorCopyFromBuffer in the SetInputTensorData method.

Thanks in advance

[asus4]

I've never tried any audio models with Unity. In general

• Use CPU mode, not GPU
• Please check whether the input type is correct. float array or quantized int16 array. Failed TensorflowLite Operation #7 this thread might be your help :)

[manena]

Netron was very useful! Thanks a lot for your prompt answer!