Unfortunately, my algorithm did not work very well on other test samples. It only correctly identified 61% of the solo singing and 82% of the non-solo singing across all seven new music selections. It looks as though the classifier did a particularly bad job with test sample 6.
Christine Smit