A novel audio-to-score alignment method using velocity-driven DTW

The problem of automatic audio to score alignment is nowadays well
understood, leading to high accuracies even for polyphonic music signals with
several instruments playing at the same time. Traditionally, the evaluation
metrics rely on the distance between the ground truth and the estimated
note onsets, considering a fixed tolerance threshold (e.g. 200 ms). This
criterion is suitable for many applications such as page turning or informed
sound source separation. However, other applications as automatic musical
accompaniment require more advance alignment, considering musical aspects
such as tempo in order to control the time differences between onsets of
consecutive notes, leading to aligned scores with more musical meaning.
The aim of this master thesis is to provide a solution to guide the align-
ment process from a more musical point of view in order to implement an
automatic musical accompaniment system. To this end, the most reliable
score follower algorithm of the MIREX competition is used as a base line
and extended taking into account the musical restrictions. The proposed
method is based on Dynamic Time Warping (DTW), where a tempo con-
troller is enforced within the minimum cost path computation in a novel
way.