next up previous contents
Next: Adaptation to polyphonic music Up: Untreated cases Previous: First and last notes   Contents

Multiple and consecutive errors

Making more than three consecutive wrong notes often leads to a bad detection or classification. This is mainly due to alignment's hesitation to find the best way in the model. To avoid this, we would have to extend our detection mechanism in order to have a greater working window, and only consider the last alignment as the right one.



Mathieu Gilles (Betr. soltau) 2003-08-25