TEMPLATE BASED KEY FINDING FROM AUDIO
Özgür İzmirli · University of Michigan Library Repository · 2005
A model for template based key finding from audio is presented and two methods that implement this model are compared.Templates are computed from a weighted combination of spectra obtained from sound recordings of monophonic musical notes.Individual weights of notes contributing to the templates are determined by profiles representing tonal hierarchies in Western music.Key determination is based on the correlations between spectral summary information obtained from audio input and the precomputed templates.The first method that implements the template based model utilizes a pure spectral representation and the second uses a chroma based representation.An overall evaluation of the model and comparison between the two methods are shown using a test audio collection.Performance results are presented for different profiles and a variety of analysis durations.Results are encouraging and show that template based models are viable for key finding from audio.