Tutorial 2: Creating an audio corpus

A Corpus represents a collection of audio sources that can be used to re-construct a given audio target. There are different ways in which we can specify the audio source(s):

• A path to a .wav, .aif, or .mp3 audio file, as a str.

• A path to a folder containing audio files, as a str.

• A list of paths to folders and/or audio files.

Note

When specifying a folder, GAMuT will recursively look for any .wav, .aif, or .mp3 audio files in the directory, ignoring any other files.

Note that creating a Corpus can easily be the most computationally expensive part when doing audio musaicking, especially when we’re dealing with large amounts of audio files. To make sure we can reuse every corpus we build, it’s best to write it to disk as a .gamut file.

Here’s a simple script showing how to write a Corpus instance to disk, as a .gamut file.

from gamut.features import Corpus

# 1) create a corpus from one or more audio sources:
# path to directory or audio file(s) to build corpus from (a list of paths is also possible)
source = [
    '/path/to/source/audio/folder-or-file-1',
    ...
    '/path/to/source/audio/folder-or-file-N',
]
corpus = Corpus(source=source)

# 2) write corpus to disk as a .gamut file
corpus.write('path/to/my_corpus.gamut')

Note

.gamut is a custom binary data format used for storing GAMuT corpora and recipe data. This file is simply a renaming of Numpy’s .npy file extension.

Next time we want to use this corpus, we simply read it from disk:

from gamut.features import Corpus

corpus = Corpus().read('path/to/my_corpus.gamut')