The following melodies have been generated by sampling the algorithms described in the paper Predictive Models for Music, in Connection Science.
The accompaniments are given (as chord progressions). The first half of each melody is given. Both algorithms try to generate the second half of the melody given the chords and the first half. The rhythms for the second halves are generated by the algorithm described in Section 2. The same rhythms and accompaniments are given to both algorithms to allow fair comparison.
The second half of each melody is beginning with a strong "beep" sound. Hence, the listener can easily know when melodies start to be generated by the algorithms.
It is obvious when listening to these examples that the melodies generated by the model constrained with Narmour features are much more realistic than the melodies generated with the simpler IOHMM.