Tests on Borodin's Polvtsian dance give the following results:
a = 76.7% of played errors have been detected.
b = 76.7% of played errors have been detected and well classified.
c = 2.8% of good notes have been recognised as errors.
Here is what we obtain with Monti's Czardas:
a = 92.2% of played errors have been detected.
b = 78.1% of played errors have been detected and well classified.
c = 4.3% of good notes have been recognised as errors.
Results for these ``long pieces'' are not very different from first results. That is a good point: size of pieces don't seem to have any influence on detection rates. It is anyway a bit amazing to see that error detection is more performant on a fast and complex piece than on a slow one. This can be explained by testing conditions: more note errors have been tested for the Czardas, which give high detection rate (a) and lower classification rate (b).
