Difference between revisions of "2007:Score File Format"
(note on xml) |
IMIRSELBot (talk | contribs) m (Bot: Fixing image links) |
||
(10 intermediate revisions by 2 users not shown) | |||
Line 1: | Line 1: | ||
+ | This page describes the proposed score file format for the MIREX 2007 | ||
+ | [[2007:Real-time Audio to Score Alignment (a.k.a Score Following)]] task. | ||
+ | |||
= Description = | = Description = | ||
The proposed MIREX score file format is in ASCII text, | The proposed MIREX score file format is in ASCII text, | ||
with one line per note or other event such as trill, tempo change, or signature change. | with one line per note or other event such as trill, tempo change, or signature change. | ||
+ | Lines are separated by newline characters and contain 10 fields, separated by whitespace. | ||
Notes are time-ordered, tempo or signature changes can come at any place. | Notes are time-ordered, tempo or signature changes can come at any place. | ||
Line 9: | Line 13: | ||
The MIREX text score file format will be also used as alignment reference format, where the clock time is no longer the score time but the aligned time in the performance, and maybe as alignment result file. | The MIREX text score file format will be also used as alignment reference format, where the clock time is no longer the score time but the aligned time in the performance, and maybe as alignment result file. | ||
− | Note that we introduced an event ID | + | === Columns === |
+ | |||
+ | Note that we introduced an field ''event ID'', that unambiguously links events across all three types of files. | ||
+ | |||
+ | The last field ''stream-id'' can serve later to separate different channels, voices, or streams of events. | ||
+ | |||
+ | The 9th field ''cue'' serves to mark events that are musically important, e.g. because they synchronise accompaniment with the performance. This could later be used for a more detailed evaluation. | ||
+ | |||
+ | |||
+ | === Example Score Files === | ||
− | + | Here is a zip with a few examples. Please have a look at them and use them to test your parsers. The corresponding midi files are in there, also, except for Anthemes 2, but that one has a trill... | |
− | + | The metric positions are not very pretty, but it's the best a simple algorithm can do. The absolute time positions are good, anyway. | |
− | |||
+ | [[Image:2007_mirex2007-score-examples.zip]] | ||
+ | === Note on higher-level formats === | ||
+ | |||
+ | It would of course be great if there was an even better format, maybe XML-based, such as the [http://www.mx.dico.unimi.it MX format] that was mentioned on the list. From this format, the text-based format could be easily generated, and even MIDI for those participants who couldn't otherwise implement a parser. | ||
+ | We'd be glad to hear a concrete proposal. | ||
= Event Types= | = Event Types= | ||
Line 22: | Line 39: | ||
== NOTE EVENTS == | == NOTE EVENTS == | ||
− | + | ==== Template ==== | |
− | event-id onset-position onset-ms type pitch interval duration-beat duration-ms cue-num stream-id | + | ''event-id onset-position onset-ms type pitch interval duration-beat duration-ms cue-num stream-id'' |
− | + | ==== Example ==== | |
<pre> | <pre> | ||
1 1 0 note 72 0 2 4000 1 0 | 1 1 0 note 72 0 2 4000 1 0 | ||
Line 34: | Line 51: | ||
</pre> | </pre> | ||
− | + | ==== Columns for note events ==== | |
# event ID [int > 1] | # event ID [int > 1] | ||
Line 50: | Line 67: | ||
== TEMPO CHANGES == | == TEMPO CHANGES == | ||
− | + | ==== Template ==== | |
− | '''0''' ''onset-position onset-ms'' '''tempo''' ''tempo - - - - stream-id '' | + | '''0''' ''onset-position onset-ms'' '''tempo''' ''tempo-bpm - - - - stream-id '' |
− | + | ==== Example ==== | |
<pre> | <pre> | ||
0 1 0 tempo 120 - - - - 0 | 0 1 0 tempo 120 - - - - 0 | ||
</pre> | </pre> | ||
− | + | This signifies a tempo of 120 beats per minute. | |
+ | |||
+ | ==== Columns for tempo changes ==== | ||
# event ID [constant int = 0] | # event ID [constant int = 0] | ||
Line 72: | Line 91: | ||
# stream ID [int] | # stream ID [int] | ||
+ | == METER CHANGES == | ||
− | == | + | ==== Template ==== |
− | + | '''0''' ''onset-position onset-ms'' '''meter''' ''numerator denominator - - - stream-id '' | |
− | + | ==== Example ==== | |
− | |||
− | |||
<pre> | <pre> | ||
− | 0 1 0 | + | 0 1 0 meter 4 4 - - - 0 |
</pre> | </pre> | ||
− | + | This signifies a rhythmic signature of 4/4. | |
+ | |||
+ | ==== Columns for signature changes ==== | ||
# event ID [constant int = 0] | # event ID [constant int = 0] | ||
# event onset position [measure+rational] (ex. 42+1/4, 28+3/7) | # event onset position [measure+rational] (ex. 42+1/4, 28+3/7) | ||
# event onset clock time [ms] (must be consistent with the onset position) | # event onset clock time [ms] (must be consistent with the onset position) | ||
− | # event type [constant symbol = ''' | + | # event type [constant symbol = '''meter'''] |
− | # | + | # meter numerator [int] |
− | # | + | # meter denominator [int] |
# unused | # unused | ||
# unused | # unused |
Latest revision as of 21:23, 13 May 2010
This page describes the proposed score file format for the MIREX 2007 2007:Real-time Audio to Score Alignment (a.k.a Score Following) task.
Description
The proposed MIREX score file format is in ASCII text, with one line per note or other event such as trill, tempo change, or signature change. Lines are separated by newline characters and contain 10 fields, separated by whitespace.
Notes are time-ordered, tempo or signature changes can come at any place. (This could be restrained to "must come at the beginning of the file", if this helps someone.)
The MIREX text score file format will be also used as alignment reference format, where the clock time is no longer the score time but the aligned time in the performance, and maybe as alignment result file.
Columns
Note that we introduced an field event ID, that unambiguously links events across all three types of files.
The last field stream-id can serve later to separate different channels, voices, or streams of events.
The 9th field cue serves to mark events that are musically important, e.g. because they synchronise accompaniment with the performance. This could later be used for a more detailed evaluation.
Example Score Files
Here is a zip with a few examples. Please have a look at them and use them to test your parsers. The corresponding midi files are in there, also, except for Anthemes 2, but that one has a trill...
The metric positions are not very pretty, but it's the best a simple algorithm can do. The absolute time positions are good, anyway.
File:2007 mirex2007-score-examples.zip
Note on higher-level formats
It would of course be great if there was an even better format, maybe XML-based, such as the MX format that was mentioned on the list. From this format, the text-based format could be easily generated, and even MIDI for those participants who couldn't otherwise implement a parser. We'd be glad to hear a concrete proposal.
Event Types
NOTE EVENTS
Template
event-id onset-position onset-ms type pitch interval duration-beat duration-ms cue-num stream-id
Example
1 1 0 note 72 0 2 4000 1 0 2 3+1/4 4500 note 60 0 0+1/4 500 2 0 3 3+3/4 5000 note 58 0 0+1/2 1000 3 0 4 3+3/4 5000 note 48 0 0+1/2 1000 0 0
Columns for note events
- event ID [int > 1]
- event onset position [measure+rational] (ex. 42+1/4, 28+3/7)
- event onset clock time [ms] (must be consistent with the onset position)
- event type [symbol: note, trill, tremolo, ...] (for tempo and signature, see below)
- event pitch [float MIDI note number]
- event interval [float halftones] (0 for non-trill)
- event duration [measure+rational]
- event duration [ms]
- cue number [int > 0, 0 = no cue]
- stream ID [int]
TEMPO CHANGES
Template
0 onset-position onset-ms tempo tempo-bpm - - - - stream-id
Example
0 1 0 tempo 120 - - - - 0
This signifies a tempo of 120 beats per minute.
Columns for tempo changes
- event ID [constant int = 0]
- event onset position [measure+rational] (ex. 42+1/4, 28+3/7)
- event onset clock time [ms] (must be consistent with the onset position)
- event type [constant symbol = tempo]
- tempo [float bpm]
- unused
- unused
- unused
- unused
- stream ID [int]
METER CHANGES
Template
0 onset-position onset-ms meter numerator denominator - - - stream-id
Example
0 1 0 meter 4 4 - - - 0
This signifies a rhythmic signature of 4/4.
Columns for signature changes
- event ID [constant int = 0]
- event onset position [measure+rational] (ex. 42+1/4, 28+3/7)
- event onset clock time [ms] (must be consistent with the onset position)
- event type [constant symbol = meter]
- meter numerator [int]
- meter denominator [int]
- unused
- unused
- unused
- stream ID [int]
Evaluation Metrics
times:
- t_a reference alignment time
- t_r reporting time
- t_e estimated time
measures:
- d_l = t_r - t_e system latency (between reporting and estimation)
- d_o = t_a - t_r offset or lag between reporting time and reference
- d_e = t_a - t_e error between estimation time and reference