Computational Music and Sound Archiving¶

The Computational Music and Sound Archiving system provides high-level audio feature extraction facilities for multi-viewpoint music similarity analysis.

Music similarity is hard to analyse. A viewpoint highlights certain aspects of musical perception. Asking for similarity regarding pitch requires another viewpoint than asking for rhythm similarity.

COMSAR combines pre-selected low-level audio features to Track objects, which represent a viewpoint.

Installation¶

Installation from PyPi¶

comsar is available on PyPi. Simply run the following command in your favorite terminal emulator:

pip install comsar

Installation from source¶

Installation from source is done in two steps:

If you do not have git installed, simply navigate to the source code repository, click on the green “Code” button and then select “Download ZIP”. Otherwise, clone the source code with git:
```
git clone https://github.com/ifsm/comsar
```
Once the code is downloaded, change to the comsar root directory and advise Python to install the pacakge:
```
cd path/to/comsar
python3 -m pip install .
```

Basic usage¶

In order to compute audio features regarding a certain track, you just have to create an instance of your desired track object and then call its extract() method with the path to an audio file. Considre the following example:

from comsar.tracks import TimbreTrack

tt = TimbreTrack()
res = tt.extract('path/to/my_audio.wav')
res.to_pickle('my_features.pkl')

The first line imports the desired Track object, in this case a TimbreTrack. The third line creates a TimbreTrack instance with the name tt. The fourth line calls the extract method of tt and passes it the path to an actual audio file. comsar then processes the audio file and makes the results available under the name res. The fifths line eventually saves the results to disc.

Track sytem¶

PitchTrack¶

RhythmTrack¶

TimbreTrack¶

class comsar.tracks.TimbreTrack(stft_params: Optional[apollon.signal.container.StftParams] = None, corr_dim_params: Optional[apollon.signal.container.CorrDimParams] = None)¶

High-level interface for timbre feature extraction.

__init__(stft_params: Optional[apollon.signal.container.StftParams] = None, corr_dim_params: Optional[apollon.signal.container.CorrDimParams] = None) → None¶

Parameters

stft_params – Parameter for STFT.
corr_dim_params – Parameter set for correlation dimension.

extract(path) → pandas.core.frame.DataFrame¶

Run TimbreTrack on audio file.

Parameters: path – Path to audio file.
Returns: Extracted features.

property n_features¶: Returns: Number of features on TimbreTrack.

FormTrack¶

Implementation of the FormTrack is planed.

Open source¶

comsar is an open source project. It is published under the permissive BSD 3-Clause License. You may change and republish the code for any personal or commercial project. The comsar source code is available on GitHub.