This is prediction demo of TensorFlow YamNet model on Unity Barracuda. YAMNet is a deep net that predicts 521 audio event classes from the AudioSet-YouTube corpus it was trained on. It employs the Mobilenet_v1 depthwise-separable convolution architecture.
The original YAMNet is composed of mel-scaled spectrogram transform and Mobilenet_v1. Because Barracuda doesn't support mel-scaled spectrogram transform. The included ONNX file only contains Moblenet_v1 and mel-scaled spectrogram transform is implemented by C#.
