The bad and the good singer: query tuning analysis for audio to audio Query by Humming

A tuning quality evaluation algorithm meant to be used as a user advising subsystem for Query-by-Humming applications is designed and presented. A queries dataset from a previous study is extended and analysed in order to define some relevant tuning quality descriptors. The automatic procedures needed to calculate those descriptors are also devised and optimized. A preexisting tuning quality expert annotation is expanded by collecting new evaluations in a specific experiment. All the available data and methods are extensively tested using machine learning software.