[en]
The DYCI2 library was designed for improvisation and style imitation: it works on sequences of abstract labels (e.g. chord labels, etc.) matching segments of a memory. DYCI2 produces new sequences by recombinations of these memory segments, given a scenario expressed as a sequence of labels, and keeping some properties learned from the memory.
The main object in om-dyci2 is DYCI2GENERATOR. It is initialized with a memory (list of "segments") and a sequence of labels indexing this memory (=> two lists of the same length).
Currently DYCI2GENERATOR only support strings as memory segments and labels.
In the simple example below the memory is a simple list of 9 segments ("s1", ..., "s9") indexed by an alphabet of 3 labels ("A", "B", or "C"). Once the generator is initialized, a query can be made using the dyi2query function, and a new list of labels driving the sequence called the scenario.
The patch dyci2generator-basic provided with the library reproduces a similar example.
In most case with real musical sequences, the memory will simply consist in indices or time markers indicating the beginning and end time of each segment. These markers can correspond to the markers in an audio file, chords or other events in a MIDI sequence, etc. (see examples below).
Depending on the application at hand, a first trick will be to convert both the list of time markers and the corresponding labels into lists of strings suitablme to initalize a DYCI2Generator. A second trick will be to recombine a sequence from the new list of memory segments.
Two cases are covered in the library examples:
- a segmented and labelled audio file
- a MIDI file including text labels.
Let's start with an OM audio file (SOUND object) containing labelled markers.
In o7, markers and labels can be set and extracted using the :markers inputs of the SOUND box. Labeled markers are output as a list of ((time "label") (time "label") ...).
From this list needs to be gereated:
- A list of segments formatted as strings: ("(t1 t2)" "(t2 t3)" ...)
- A list of labels
The result of a query is a new list of ("(ta ta+1)" ... "(tx tx+1)" ...). From this list needs to be retrieved and sequenced segments if the original audio file. This can also be done using standard OM audio tools.
=> See the patch dyci2generator-audio provided in the library
OM/o7 provide tools to process files and extract labels and time markers coming from external tools.
IRCAM's AudioSculpt for instance provides powerful means to automatically segment an audio file in beats and let you input labels attached to the beat markers. Markers can then be exported as SDIF files, and imported/processed in OM:
Depending on your annotation and storage convention, different patches will need to be programmed to retreive the correct format for memory and labels.
The function dyci2query allows to get a new generated sequence from a trained DYCI2GENERATOR object and new list of labels (or scenario).
Note: the labels in the scenario must all be part of the original training labels.
Depending on how the memory was formatted, adequate operations must be done to reconstitute a musical sequence (e.g. concatenating MIDI chucks, sequencing audio buffers, etc.)
The TEXTBUFFER object can be convenient to easily input queries in dyci2query. Use the :read-mode input's option to get the adequate contents format (e.g. using the "flat-list" option), and convert to string to match the DYCI2GENERATOR labels' format.
=> Avoid space in labels, which might be wrongly parsed and processed.
We also have a lot of options with which you can drive your scenario generation: with the dyci2setparam command, you can connect on the SELF spot an element from the DYCI2 Generator class, on the TYPE spot type of your input (currently, only integers exist in), on the PARAMETER spot, the parameter you want to change (currently, only "avoid_repetitions_mode" and "max_continuity" exist) and in the fourth spot, the VALUE spot, put the value of your parameter.
-
avoid\_repetitions\_mode:- set at 0: it avoids many use of the same segment from the memory in the output sequence.
- set at 1: it favors the least used segments for the output sequence.
- set at 2: it allows repetitions of every segments in the output sequence.
-
max\_continuity: the input must be a natural integer which represents the maximum length of a segments sequence from the memory in the output sequence.