Introduction

The objective of the survey is to evaluate the performance of a harmonic progression generator. The survey was created based on the framework described by M. Pearse in Towards a framework for the evaluation of machine compositions. As stated in the framework, there are some aspect of the generation that should be clear before completing the survey:

• The aim of the proposed system is to generate harmonic progression based on a particular style. The style will be defined by the data corpus used in the system training.
• The system relies on probabilistic models for the training and generation.
• The training data corpus used for the traing was McGill Billboard Project.
• The generated samples used for the present survey were selected randomly from a set of generated samples.
• The human generated samples used for the present survey were selected randomly from the complete training set.
• Given that both input and output of the proposed system are symbolic chord progression, we needed to convert this is to musical scores for the proper evaluation. Then, following samples were composed by the same human who was given 9 harmonic progression. The human composer was the same and the method used was the same (4-voice compositions). The human composer was not aware about which progression corresponded to the training set or automatically generated set.

The evaluation will be done by comparing the similarities and diferences between convetional melodies that we used as a training data and the ones composed by the system proposed.

Three groups of three melodies will be presented. One melody in each group has been composed by our chord progression generator. The two other melodies are real chord progression created by human composers and included in the traning data set.

Please listen to them carefully and identify, using the buttons provided, the progression that, in your opinion, was generated by our automatic system. Note that and extra option is available (I don't know) that should be chosen if you can not identify the automatically generated progression.

Next, we would like you to choose the one, on each group, you liked the most.

(Questions without * are optional.)