Results 1 
5 of
5
Pitch Spelling Algorithms
, 2003
"... In this paper I introduce a new algorithm called ps13 that reliably computes the correct pitch names (e.g., C#4, B#5 etc.) of the notes in a passage of tonal music, when given only the onsettime and MIDI note number of each note in the passage. ps13 correctly predicts the pitch names of 99.81% of t ..."
Abstract

Cited by 36 (11 self)
 Add to MetaCart
In this paper I introduce a new algorithm called ps13 that reliably computes the correct pitch names (e.g., C#4, B#5 etc.) of the notes in a passage of tonal music, when given only the onsettime and MIDI note number of each note in the passage. ps13 correctly predicts the pitch names of 99.81% of the notes in a test corpus containing 41544 notes and consisting of all the pieces in the first book of J. S. Bach's Das Wohltemperirte Klavier (BWV 846869). Three previous algorithms (those of Cambouropoulos (1996, 1998, 2002), LonguetHiggins (1987) and Temperley (2001)) were run on the same corpus of 41544 notes. On this corpus, Cambouropoulos's algorithm made 2599 mistakes, LonguetHiggins 's algorithm made 265 mistakes and Temperley's algorithm made 122 mistakes. As ps13 made only 81 mistakes on the same corpus, this suggests that ps13 may be more robust than previous algorithms that attempt to perform the same task.
Separating voices in polyphonic music: A contig mapping approach
 In Computer Music Modeling and Retrieval: Second International Symposium
, 2004
"... Abstract. Voice separation is a critical component of music information retrieval, music analysis and automated transcription systems. We present a contig mapping approach to voice separation based on perceptual principles. The algorithm runs in O(n 2) time, uses only pitch height and event boundari ..."
Abstract

Cited by 23 (0 self)
 Add to MetaCart
(Show Context)
Abstract. Voice separation is a critical component of music information retrieval, music analysis and automated transcription systems. We present a contig mapping approach to voice separation based on perceptual principles. The algorithm runs in O(n 2) time, uses only pitch height and event boundaries, and requires no userdefined parameters. The method segments a piece into contigs according to voice count, then reconnects fragments in adjacent contigs using a shortest distance strategy. The order of connection is by distance from maximal voice contigs, where the voice ordering is known. This contigmapping algorithm has been implemented in VoSA, a Javabased voice separation analyzer software. The algorithm performed well when applied to J. S. Bach’s Twoand ThreePart Inventions and the fortyeight Fugues from the WellTempered Clavier. We report an overall average fragment consistency of 99.75%, correct fragment connection rate of 94.50 % and average voice consistency of 88.98%, metrics which we propose to measure voice separation performance. 1
(a) (b)
"... ifH [i +1][n 2 ][2] = H [i][n 1 ][2] 9if(H [i +1][n 2 ][1] (H [i +1][n 2 ][2] + 1) mod 7 11 if (H [i][n 1 ][1] H [i +1][n 2 ][1]) mod 12 (H [i +1][n 2 ][2] Figure 8: Algorithm for correcting neighbour note errors. 9/13 4ifH [i +2][n 3 ][2] = (H [i][n 1 ][2] ..."
Abstract
 Add to MetaCart
ifH [i +1][n 2 ][2] = H [i][n 1 ][2] 9if(H [i +1][n 2 ][1] (H [i +1][n 2 ][2] + 1) mod 7 11 if (H [i][n 1 ][1] H [i +1][n 2 ][1]) mod 12 (H [i +1][n 2 ][2] Figure 8: Algorithm for correcting neighbour note errors. 9/13 4ifH [i +2][n 3 ][2] = (H [i][n 1 ][2] 7if0 < (H [i][n 1 ][1] [i +1][n 2 ][1]) mod 12 < (H [i][n 1 ][1] [i +2][n 3 ][1]) mod 12 10 if H [i +1][j][2] = (H [i][n 1 ][2] (H [i][n 1 ][2] Figure 9: Algorithm for correctingdescending passing note errors. 10/13 4ifH [i +2][n 3 ][2] = (H [i][n 1 ][2] + 2) mod 7 7if0 < (H [i +2][n 3 ][1] [i +1][n 2 ][1]) mod 12 < (H [i +2][n 3 ][1] 10 if H [i +1][j][2] = (H [i][n 1 ][2] + 1) mod 7 (H [i][n 1 ][2] + 1) mod 7 Figure 10: Algorithm for correctingascending passing note errors. 11/13 O # [i] O 1 M # ##O # [i][2]# 7 return M # Figure 11: Algorithm
PITCH SPELLING WITH CONDITIONALLY INDEPENDENT VOICES
"... We introduce a new approach for pitch spelling from MIDI data based on a probabilistic model. The model uses a hidden sequence of variables, one for each measure, describing the local key of the music. The spellings in the voices evolve as conditionally independent Markov chains, given the hidden ke ..."
Abstract
 Add to MetaCart
(Show Context)
We introduce a new approach for pitch spelling from MIDI data based on a probabilistic model. The model uses a hidden sequence of variables, one for each measure, describing the local key of the music. The spellings in the voices evolve as conditionally independent Markov chains, given the hidden keys. The model represents both vertical relations through the shared key and horizontal voiceleading relations through the explicit Markov models for the voices. This conditionally independent voice model leads to an efficient dynamic programming algorithm for finding the most likely configuration of hidden variables — spellings and harmonic sequence. The model is also straightforward to train from unlabeled data, though we have not been able to demonstrate any improvement in performance due to training. Our results compare favorably with others when tested on Meredith’s corpus, designed specifically for this problem. 1