Yamnet is a audio classification NN based on Mobilenet V1 architecture. The NN has been trained on Audio Set Dataset which contains 527 classes of audio events.
This project contains an implementation on GAP processor of YAMNet specialized for inference on mobile devices with the following changes:
- The model is quantized. This required retraining the model with Relu6 non-linearities instead of Relu, to limit activation ranges.
- The inference signature is simpler. We now take a single fixed-length frame of audio (15600 samples) and return a single vector of scores for 521 audio event classes.
Please refer to the original YAMNet for more information about the model description, as well as suitable uses and limitations.
TFLITE reference Project: Tensor Flow Hub project
The model accepts a 1-D float32 Tensor or NumPy array of length 15600 containing a 0.975 second waveform represented as mono 16 kHz samples in the range [-1.0, +1.0].
The model returns a 2-D float32 Tensor of shape (1, 521) containing the predicted scores for each of the 521 classes in the AudioSet ontology that are supported by YAMNet. The column index (0-520) of the scores tensor is mapped to the corresponding AudioSet class name using the YAMNet Class Map, which is available as an associated file yamnet_label_list.txt packed into the model file. See below for usage.