MIDI Database for Content-Based Retrieval

Name

Synopsis

Description

First, an index must be created. The following creates an index called lead.mdb using 10 pitch directions and secondary contours.

mididb -f lead.mdb -k 10 -x create

Files are then added to the index. It is possible to use a classifier to select parts of the file to add. Here, the lead parts of file.mid, as determined by the hybrid classifier, are added to the index.

mididb -f lead.mdb -c hybrid -p lead add file.mid

The index can then be queried. The following example uses a midi file with the secondary contour (which gives more specific results but requires a more accurate query).

mididb -f lead.mdb -x findmidi query.mid

Options

-a Use summary file (for analysis functions)

-b time The start time (in milliseconds) of the segment to use. Does not apply to all commands Default: 'start of the file'

-c classifier The part classifier to use. Default: hybrid, other options are: stats; net and knearest

-d factory The database factory to use (e.g. onix ).

-e time The end time (in milliseconds) of the segment to use. Does not apply to all commands. Default: 'end of the file'

-f file Specify a different database filename. Has no effect on non-DB operations. Default: sbh

-g distance Specify how similar a contour must be over the entire query. Default: 2

-h Displays help similar to this page.

-i Perform validation of search results (if useful).

-k size Define the number of pitch directions per atomic unit. Note - this only applies to db creation, 'lc' and 'qeg'. Default: 12

-l distance Specify how similar a contour must be per atomic unit. Default: 2

-m mapping Specify the contour map to use. Available maps are: dowling; extended; alphabet. Default: dowling

-n limit Define the limit on near match set for contours. Default: 20000

-p part Specify a part of the midi file to use in the operation. Valid parts are: acc; bass; drums and lead. Default is to use everthing but drums.

-q nths Specify the 1/nths of a note to quantise input midi to. Note - set to 0 to prevent quantisation. Default: 32

-r limit Define the limit on the number of matches returned. Note - this applies to DB search and time indexing. Default: 50

-v value Specify a value for operations which require parameters.

-x Enable extended functionality of some commands.

Commands

[add | a] MIDI_file

Add a MIDI file to the database.

[addlist | al] List_file

Add a list of MIDI files to the database.

[makelink | ml] MIDI_file Destination

Add a link to the database.

[create | cr]

Create a database.

[find | f] contour

Search the database for the contour.

[findmidi | fm] MIDI_file

Search the database using a MIDI file as the query.

[follow | fo] contour

Search the database links pertaining to the contour.

[calcexpand | ce]

Display the estimates for the sizes of expanded queries. The -v option specifies the size of the query to consider.

[classify | cl] MIDI_file

Displays the part classification of the tracks in the file.

[contours | co] MIDI_file

Displays full file info - including contours. The -b and -e options can be used to specify a segment of the file.

[nearmatch | nm] contour

Displays the near match set for this contour.

[savemidi | sm] Input_MIDI_file Output_MIDI_file

Performs all standard processing on the input and then saves it. This is primarily useful as a debug tool - to be sure of what is being used to query or add to the database.

[showparams | sp]

Displays the architecture of the DB.

[tempomap | tm] MIDI_file

Displays the tempo map for the MIDI file.

[timeindex | ti] MIDI_file contour

Displays the time indexes of contours in the MIDI file.

[timeindexmidi | tim] MIDI_file contour

Same as timeindex, using a MIDI file as the query.

defrag destination-file

Defragment chains in the database file. Also currently has no standard feedback to user (just sits there).

bottom output_file

Outputs the least popular contours based on hits and chains. The -v switch defines the size of the chart to compile.

[distgraph | dg] output_file

Outputs stats, suitable for graphing, which shows the distribution of matches across the database. The -v switch gives the resolution of the graph.

[freqgraph | fg] output_file

Outputs stats, suitable for graphing, which shows the frequency of database matches and uniqueness of contours.

[ffreqgraph | ffg] output_file

Similar to freqgraph but simulates query expansion. The -v switch defines the percentage (1-100) of the database to sample.

[listcontours | lc]

Displays a list of all possible contours (for the given atomic length).

[mksummary | ms]

Makes a summary file which can be used for some analysis functions.

top output_file

Outputs the most popular contours based on hits and chains. The -v switch defines the size of the chart to compile.

[qexpgraph | qeg] output_file

Analyse the size of expanded queries (for the given atomic length).

[accuracy | acc] query_file output_file

Evaluate the accuracy of the database. Uses the current search parameters.

[describe | desc] input_file output_file

Describe a set of midi files, as listed in the input file.

[evalqexp | eqx] output_file

Evaluate the time taken to perform query expansion (for the given atomic length). The -v switch defines the time limit for the test (in minutes). The -x switch indicates that brute force matching should be used.

[timefind | tf] output_file

Evaluate the time taken to perform a search (uses the given database). The -v switch defines the time limit for the test (in minutes).

Remarks:

Implemented by midi_db_main.cpp.