Comparison of approaches for instrumentally predicting the quality of text-to-speech systems