What is it?

AudioClassifier is an openFrameworks application, part of the ml4a-ofx collection, which can be trained to distinguish different sounds from each other and output the result to other applications over OSC.

AudioClassifier has been used by Støj to play Wolfenstein 3D only using sounds.

Training instructions

  1. Press the record button to start/stop recording training examples on the current class.

  2. Switch the class label by pulling the Class Label slider and record new training examples

  3. Repeat step 2 for however many distinct sounds you intend to classify.

  4. Press Train to train the model

  5. Make some sounds and see how they get classified

The video below takes you through the steps of the training process.

Threshold mode

When Threshold Mode is on, the app will only react to sounds louder (RMS) than volThreshold.

Additionally the Threshold timer (ms) will block any new inputs for a number of milliseconds.

Threshold mode effects both when training new examples and classifing new inputs.

OSC Output

By default the app outputs OSC to localhost, port 8000, adress “/classification”. This can be changed in ofApp.h.

Key inputs

As an alternative to using the GUI you can use the following key inputs:

Setup and training considerations