Results

Winners

The winner in the automatic category was

A stochastic model of artistic deviation and its musical score for the elucidation of performance expression, Kenta Okumura, Shinji Sako, and Tadashi Kitamura, Nagoya Institute of Technology, Japan

The winner in the semi-automatic and interactive category was

VirtualPhilharmony: a conducting system focused on a sensation of conducting a real orchestra, Takashi Baba (1), Mitsuyo Hashida (2), Haruhiro Katayose (1), (1) Kwansei Gakuin University, (2) Soai University

Procedure

The two obligatory pieces that were used were the first parts of
1. Sonata K.466 by Domenico Scarlatti (MIDI: 1_Scarlatti_K466)
2. Prelude No. 3 by Nino Rota (MIDI: 2_Rota_Prelude_No3)
The were handed out to the participants about two hours before the competition started.

The Rencon competition had a relatively large attendance with a total number of 80 questionnaires that were filled in by the audience.
The jury consisted of the two Swedish concert pianists
Lucia Negro and Stefan Bojsten.

The questionnaire had four rating scales for each example ranging from 1 to 7. It was “Technical control”, Musicality”, “Humanness”, and “Expressive variation” (see call for participation). The four ratings were averaged to obtain the final value. These sums were then averaged across audience and jury separately.

All the performances that were computed in advance were played anonymously in a random order. In addition, the nominal version (without any performance variations) of each song was played as the first two examples. The interactive versions were performed live at the end of the listening session.

Due to the rather small number of participants it was decided in beforehand to merge the interactive and semi-automatic category into a category labelled “human intervention”.

Detailed results

The averaged ratings for each system divided according to composer and judgement group are shown in the table below. The audio examples were recorded at the competition using the same recording volume across all examples.

Scarlatti Rota Total Rank Scarlatti Rota
Audience Jury Audience Jury Audio Audio
Nominal 3.6 2.4 3.3 2.9 3.0  ex1-scar-nom  ex2-rota-nom
Automatic
Polyhymnia+ 4.1 2.8 4.5 2.4 3.4 2  ex3-scar-poly  ex6-rota-poly
Stochastic model 3.9 2.4 4.3 4.0 3.6 1  ex7-scar-stoc  ex10-rota-stoc
Kagurame 4.1 3.0 2.8 1.6 2.9 3  ex9-scar-kagu  ex4-rota-kagu
Human intervention
Accent model 4.6 3.3 4.3 2.6 3.7 3  ex5-scar-acc  ex8-rota-acc
VirtualPhilharmony 4.8 3.6 5.1 4.1 4.4 1  ex11-scar-virt  ex12-rota-virt
CaRo 2.0 4.5 1.8 5.0 4.0 3.8 2  ex13-scar-caro  ex14-rota-caro

As seen in the table above, the overall judgements of the jury were lower than the audience ratings. However, there was a fair agreement between the two groups with a correlation between them of  r=0.66 (n=14).

The results for each rating scale averaged over the two jury members are shown below.

Scarlatti Rota
Technical Musical Human Expressive Technical Musical Human Expressive
Nominal 3.5 1.5 2 2.5 5 2 2 2.5
Automatic
Polyhymnia+ 1.5 2.5 3 4 2 2.5 2.5 2.5
Stochastic model 2.5 2 2.5 2.5 4.5 3.5 3 5
Kagurame 3.5 3.5 3.5 1.5 1 1.5 1.5 2.5
Human intervention
Accent model 4 3 3.5 2.5 2.5 2.5 3 2.5
VirtualPhilharmony 3 3 4 4.5 2.5 3.5 4.5 6
CaRo 2.0 1 1.5 2 2.5 4 3 3.5 5.5

As expected the nominal versions obtained high ratings on the “Technical” scale and low on the others. Note that the highest ratings for “Human” and “Expressive” were obtained for the VirtualPhilharmony system which was the only system that was controlled using real gestures.