Difference between revisions of "2006:Symbolic Melodic Similarity"

From MIREX Wiki
(Building the ground truth)
(Rainer Typke and Anna Pienimäki)
Line 54: Line 54:
  
 
=== Rainer Typke and Anna Pienimäki ===
 
=== Rainer Typke and Anna Pienimäki ===
* About pooling, we think that is more reasonable to use the similar procedures as last year when deciding on the amount of similarity between the melodies for technical, music philosphical and comparability reasons.
+
* About pooling, we think that is more reasonable to use procedures similar to last year when deciding on the amount of similarity between the melodies for technical, music philosphical and comparability reasons.
* We think that when evaluating the melodies, the complete melodies should be made available but also matching snippets (if the algorithm can provide them). Asking the evaluators to judge up to 5 minute songs instead of relevant snippets is not practical. For instance, if we have n = 10 returned items per participant per query and m = 10 queries and the average length of the song is k = 3 minutes and p = 4 participating algorithms, it takes m*n*k*p = 10*10*3*4 = 1200 minutes to just listen through all the candidate songs. With 10 sec snippets, we end up with just over an hour, which is more reasonable.
+
* We think that when evaluating the melodies, the complete melodies should be made available but also matching snippets (if the algorithm can provide them). Asking the evaluators to judge up to 5 minute songs instead of relevant snippets is not practical. For instance, if we have n = 10 returned items per participant per query and m = 12 queries and the average length of the song is k = 3 minutes and p = 4 participating algorithms, it takes m*n*k*p = 10*12*3*4 = 1440 minutes to just listen through all the candidate songs. With 10 sec snippets, we would end up with an hour and 20 minutes, which is more reasonable.
 
* Algorithm running time: If the IMIRSEL team can make it available, this information would be interesting:  
 
* Algorithm running time: If the IMIRSEL team can make it available, this information would be interesting:  
 
** time spent by the system
 
** time spent by the system
 
** time spent by the algorithm for indexing (if there is indexing)
 
** time spent by the algorithm for indexing (if there is indexing)
 
** time spent by the algorithm for searching
 
** time spent by the algorithm for searching

Revision as of 06:29, 13 June 2006

Overview

This page is devoted to discussions of The MIREX06 Symbolic Melodic Similarity contest. Discussions on the MIREX 06 Symbolic Melodic Similarity contest planning list will be briefly digested on this page. A full digest of the discussions is available to subscribers from the MIREX 06 Symbolic Melodic Similarity contest planning list archives.

Task suggestion: Symbolic Melodic Similarity

Proposed tasks

1. Retrieve the most similar incipits from the UK subset of the RISM A/II collection (about 15,000 incipits), given one of the incipits as a query, and rank them by melodic similarity. Both the query and the collection are monophonic. Half the queries are hummed or whistled queries that have been converted to MIDI, thus with slight rhythmic and pitch imperfections, and half the queries are quantized in pitch and rhythm.

2. Like task 1, but with two collections of mostly polyphonic MIDI files to be searched for matches. The queries would still be monophonic. The first collection would be 10,000 randomly picked MIDI files from a collection of about 60,000 MIDI files that were harvested from the Web. They include different genres (Western and Asian popular music, classical music, ringtones, just to name a few). The second collection would be more focused: about 1000 .kar files (Karaoke MIDI files) with mostly Western popular music which stem from the same web harvest.

Inputs/Outputs

Task 1: Input: Parameters: - the name of a directory containing about 15,000 MIDI files containing mostly monophonic incipits and - the name of one MIDI file containing a monophonic query.

The program will be called 6 times. Three of the queries are going to be quantized (produced from symbolic notation) and three produced by humming or whistling, thus with slight rhythmic and pitch deviations.

Expected Output: - a list of the names of the 10 most similar matching MIDI files, ordered by melodic similarity. Write the file name in separate lines, without empty lines in between.


Task 2: Input: same interface as for task 1, thus the name of the directory with files to be searched and the name of the query. However, the directory will contain either about 10,000 mostly polyphonic MIDI files or 1000 Karaoke files.

Output: a list of the names of 10 MIDI file names that contain melodically similar musical material, ordered by similarity, plus for each file the time (offset from the beginning in seconds) where the query matches and where the matching bit ends. If the query matches in more than one position, return the position of the most similar match (or any one of them if there is more than one most similar match). If the algorithm does not align the query with the MIDI file at any particular position, just return 0 as start time and the duration of the MIDI file as end time.

Building the ground truth

Unlike last year, it is now nearly impossible to manually build a proper ground truth in advance.

Because of that, after the algorithms have been submitted, their results are going to be pooled for every query, and every participant is going to be asked to judge the relevance of the matches for some queries. To make that a manageable burden, it is important that the algorithms do not only return the names of the matching MIDI files for task 2, but also where the matching bit starts and ends in the matching MIDI file. We can then automatically extract those matching bits and put them into small new MIDI files whose relevance can then be quickly checked.

Measures

Use the same measures as [last year] to compare the search results of the various algorithms.

Potential participants

The following people have confirmed their interest:

  • Klaus Frieler
  • Nicola Orio
  • Kjell Lemstr├╢m
  • Rainer Typke

Nicola Orio

  • Pooling among participants of the first M results of the retrieved files (as in the proposal), with a binary relevance, and if not binary at least quantized in a small number of classes (like 0-5 or 0-3).
  • The ability to identify the part of the melody to be presented to the evaluator can be done (and make sense) only for local alignment approaches, not for approaches on more general properties of the melodies. I'm afraid this difference in presenting the retrieval results will bias the assessments, and I suggest to use complete melodies for the pooling of the results.

Kjell Lemstr├╢m

  • When evaluating the time that algorithms use for the task, the time that is spent by system should be excluded from algorithm running times (this was not the case last year).

Rainer Typke and Anna Pienimäki

  • About pooling, we think that is more reasonable to use procedures similar to last year when deciding on the amount of similarity between the melodies for technical, music philosphical and comparability reasons.
  • We think that when evaluating the melodies, the complete melodies should be made available but also matching snippets (if the algorithm can provide them). Asking the evaluators to judge up to 5 minute songs instead of relevant snippets is not practical. For instance, if we have n = 10 returned items per participant per query and m = 12 queries and the average length of the song is k = 3 minutes and p = 4 participating algorithms, it takes m*n*k*p = 10*12*3*4 = 1440 minutes to just listen through all the candidate songs. With 10 sec snippets, we would end up with an hour and 20 minutes, which is more reasonable.
  • Algorithm running time: If the IMIRSEL team can make it available, this information would be interesting:
    • time spent by the system
    • time spent by the algorithm for indexing (if there is indexing)
    • time spent by the algorithm for searching