Measuring music transcription results based on a hybrid decay/sustain evaluation

Although much work is being done in music transcription research, the evaluation of these techniques is less addressed by the research community. The lack of widely accepted metrics and databases presents an obstacle to the assessment of existing music transcription approaches. This paper presents an analysis of existing metrics and proposes a new method for measuring the results of music transcription. Based on the idea that decay and sustained music instruments may have different requirements, a dual process is implemented. On the decay process, a note oriented approach is used, considering pitches and onsets, generating a score for each note. On the sustain process, a time oriented approach is used, measuring the overlap of original and transcribed notes. The final score is produced based on the values obtained in both processes. To evaluate the proposed approach, several music transcription metrics were compared with human tests results. The obtained results show that the proposed method achieves the best correlation with human perception results. Based on the idea that not all transcription errors have the same impact, an effort was made to achieve a metric that is more realistic from the human perception point-of-view.

[1]  Valentin Emiya,et al.  Perceptually-Based Evaluation of the Errors Usually Made When Automatically Transcribing Music , 2008, ISMIR.

[2]  Helmut Haas,et al.  The Influence of a Single Echo on the Audibility of Speech , 1972 .

[3]  Christopher Raphael,et al.  Automatic Transcription of Piano Music , 2002, ISMIR.

[4]  G. Reis,et al.  Genetic Algorithm Approach to Polyphonic Music Transcription , 2007, 2007 IEEE International Symposium on Intelligent Signal Processing.

[5]  Daniel Leech-Wilkinson,et al.  European Society for the Cognitive Sciences of Music (ESCOM) , 1991 .

[6]  M.P. Ryynanen,et al.  Polyphonic music transcription using note event modeling , 2005, IEEE Workshop on Applications of Signal Processing to Audio and Acoustics, 2005..

[7]  N. Fonseca,et al.  A Genetic Algorithm Approach with Harmonic Structure Evolution for Polyphonic Music Transcription , 2008, 2008 IEEE International Symposium on Signal Processing and Information Technology.

[8]  Daniel P. W. Ellis,et al.  A Discriminative Model for Polyphonic Piano Transcription , 2007, EURASIP J. Adv. Signal Process..