Latest 0.3.3
License Proprietary
Platforms ios 8.4, osx 10.11
Dependencies HDF5Kit, Peak/MIDI, Upsurge

Intune Features

The IntuneFeatures framework contains code to generate features from audio files and feature labels from the respective MIDI files. Currently supports these features:

  • [x] Log-scale specturm power estimate by bands
  • [x] Spectrum power flux
  • [x] Peak power
  • [x] Peak power flux
  • [x] Peak locations

The CompileFeatures command line app takes audio and MIDI files as input and generates HDF5 databases with the features and the labels. These HDF5 files can then be used to train a neural network for transcription or related tasks.


Features are extracted from each consecutive window of audio data. The window size and the step size between windows can be configured, but the window size must be a power of 2. Each window is then multiplied by the Hamming windowing function. To get the spectal power the FFT is computed for each window and the resulting spectrum is divided into bands of the equal temperament scale.

Peaks are extracted from the FFT and filtered by a height threshold and by a minimum peak distance requirement. Peak locations are computed for each band as the distance between the peak’s frequency and the band’s middle frequency.


IntuneFeatures is available under the MIT license. See the LICENSE file for more info. Copyright © 2016 Venture Media Labs.

Latest podspec

    "name": "IntuneFeatures",
    "version": "0.3.3",
    "summary": "Audio feature extraction",
    "homepage": "",
    "license": "Proprietary",
    "authors": {
        "Alejandro Isaza": "[email protected]"
    "platforms": {
        "ios": "8.4",
        "osx": "10.11"
    "source": {
        "git": "",
        "tag": "0.3.3"
    "source_files": "Sources/**/*.swift",
    "resource_bundles": {
        "NoteCurves": "Resources/note_curves.h5"
    "dependencies": {
        "HDF5Kit": [
            "~> 0.1"
        "Peak/MIDI": [
            "~> 1.2"
        "Upsurge": [
            "~> 0.7.1"

Pin It on Pinterest

Share This