A Release Notes
A.1 Version 1.0, released 05.07.2002
RELEASE NOTES FOR jMAX PACKAGE "SUIVI" FOR SCORE FOLLOWING
05.07.2002 Version 1.0
Well working versions (with some shortcomings),
help and demo patches.
All objects: The package version and build date is output on
instantiation and start messages.
1. suiviaudio
=============
Working well except repeated notes for the voice.
- Rest modeling parameters can be set while running
- Vowel change probability (from cepstral difference) is output
2. suivimidi
============
Debugged and tested, but for monophonic performances only
(polyphonic can work but is not guaranteed)
New features:
- Note-offs are taken into account for rest detection and modeling
- Trills are represented as a chord on a special midi channel, given as a
parameter
- Zigzags can be disallowed
- Debug printing, with several levels of verbosity to choose from
Fixed bugs:
- Note-offs were handled incorrectly
- Occasional crashes when the score-window ended with a rest
- Label assignment would make the follower hang on certain cues
3. Helper objects
=================
- suivieval
Quantitative evaluation of a follower, given a reference
performance in midi with the same cues as the score. Output is
the average detection latency and its standard deviation, the
number of errors (detection outside a threshold given as
parameter), and the percentage of recognised cues.
- suivimakecue
Allows to generate cues for each note in a sequence, a chord is
assigned a single cue by virtue of a time-window given as
parameter.
2000 - 04.2002 All versions before (without clear version output)
A.2 Version 1.1, released 18.09.2002
RELEASE NOTES FOR jMAX PACKAGE "SUIVI" FOR SCORE FOLLOWING
18.09.2002 Version 1.1
1. suivimidi
============
- unified polyphonic score modeling
- maxdiff parameter to set max time between notes to become a chord
- force forward/backward
- label can be assigned to rest
- polyphonic MIDI input
- weight for attack, parameter attack_weight
2. suiviaudio
=============
- unified polyphonic score modeling (with ghost states)
- maxdiff parameter to set max time between notes to become a chord
- force forward/backward
- label can be assigned to rest
- nozigzag mode
- global tuning parameter
- silence threshold parameter
- uses score representation for fricatives
(pitch 127 on special midi channel) and trills (chord on special channel)
- messages to set a single or all PDF parameters (loadable from text file)
- demos/test_enecho.jmax is an incredible patch to test all parts of enecho
3. Old Objects
==============
The objects from package version 1.0 are still available (for comparison) as
suivimidi_v1_0 and suiviaudio_v1_0.
(See demos/explo_audio_shootout.jmax for an example.)
4. Helper Objects and Templates
===============================
- suivimapper: map ranges of input labels to an output label
- template evaluation simplifies patches with suivieval
(See demos/explo_audio_shootout.jmax for an example.)
- templates click and synth factor out often used sub-patches
5. Objects in Preparation of Training
=====================================
- suivimakeref: dump model into text file
- suiviaudiofeat: output features analysed by suiviaudio,
to be written to SDIF files
A.3 Version 1.2, released 20.12.2002
RELEASE NOTES FOR jMAX PACKAGE "SUIVI" FOR SCORE FOLLOWING
20.12.2002 Version 1.2 Release Candidate 1
This is the last release under jMax 2.5, the next release will be for
jMax 4 only.
1. suivimidi
============
- debugging of score parsing for highly polyphonic scores, no more hanging
- addition of a note-match probability --> Pluton piano works
- correct distinction of silence/not silence for ghost-rest
- removed explode-emulation mode
- clearer verbose 1 output of detected state
- verbosity can be changed while running
2. suiviaudio
=============
- debugged FFT window, so old PDF parameters didn't work anymore
- integration of new PDF parameters from statistical analysis by default
- clearer verbose 1 output of detected state
- PDF parameters and verbosity can be changed while running