Next: , Previous: Command-line tools, Up: Tour


3.0.2 User interfaces

A variety of graphical user interfaces are provided with the Marsyas source distribution. Although it is possible to write a user interface that communicates with Marsyas in any language there is specific support for interfacing with the Qt toolkit by Trolltech http://www.qtsoftware.com/products/. In order to compile the graphical user interfaces you will need to have Qt4 installed and enable the WITH_QT using CMake. More information can be found at the chapter Source installation.

MarPlayer is a simple audio player that provides a seekable playback progress indicator while playing audio and showcases multi-threading using Qt4 and Marsyas.

     cd MY_MARSYAS_DIR/build/bin
     MarPlayer

This will launch the MarPlayer GUI. Click on File and open one of the audio files in the collections (or any file in a Marsyas supported format). Clicking on the playback slider will seek to the corresponding location in the audio file.

     cd MY_MARSYAS_DIR/build/bin
     MarPhasevocoder

Open a file and experiment with the sliders. The Frequency and Time sliders can be used to pitch shift the recording without changing the duration or speed up or slow down the recording without changing the pitch respectively. The Sinusoids slider can be used to control the number of sinusoids (sorted by amplitude) that are used to approximate the audio signal at each frame. This example showcases user interaction with a relatively performance intensive audio synthesis technique like the phasevocoder which frequently does not have real-time implementations.

The last example of a user interface is a content-based music browsing interface for large audio collections based on self-organizing maps. First you will need to create the genres10.mf collection file as described in Section Command-line tools.

     cd MY_MARSYAS_DIR
     MarGrid2

Click on File-Open-Open Collection File and select the genres10.mf collection. Then click on the E button (Extract) which performs feature extraction for all the 1000 files in the collection. This will take a few minutes and you can view the progress in the terminal output. When the feature extraction is complete click on the T button (Train) which trains a self-organizing map that maps the high-dimenstional continuous audio features representing each song to 2D coordinates on a grid. This takes a few seconds. Now click on the P (Predict) button to place each song on the grid. Feature extraction is performed again therefore this takes about the same time as the Train stage. Click on View-Colour-Mapping mode to see a visualization of the genre distributions over the self-organizing map. Note that the genre information is only used for display purposes but not during the calculation of the mapping. If either the audio features or the self-organizing map did not work the colors would essentially appear randomly distributed. Each square contains one more more tracks that are similar to each other based on the audio feautures. Clicking on a squares allows the user to cycle through the songs. Another interesting feature can be activated by selecting View-Continuous which switches songs continuously as the user hovers over the space without requiring explicit clicking. This mode is particularly effective when using touch surface interaction. Once a mapping is calculated it is possible to save the grid and load it without requiring the time consuming stages of feature extraction and training.