analysis and creative collaborations
clear that we have knowledge from several areas
how best to profit from each other
Interdisciplinarity/collaboration ideas/issues
real-time MIR in general MIR it's not so interesting but there are some people that will be interested
definite minority of people making real-time tools
in terms of the focus of rhythm "non-western" what is the connection? how can the two sides work on this - it's ill defined
in less experimental works, what useful things can happen
Typical problem with MIR is the awful annotations because not many musically trained people are involved in the corpus building - composers / musicians can help here
may not take that long for trained musicians - but will they want to??
World music is more interesting compositionally but MIR concerned mainly with "west"
True collaboration would lead to lots of new interesting research questions and paradigms that are useful to both communities
many people in MIR are not musically motivated - a lot of the research has no music relevance (exept maybe genre classification)
MIR is not asking some fundamental (musicological) question about the data - can we feedback some of these ideas
The fact the benefit is there does not mean we can fit an agenda - we need to narrow down and get an agenda
we cannot rely on the community to motivate themselves
Generative
coming from the "generative" side the analytical gives accurate picture of what's happening outside the software that the software can respond to as quickly as possible
how best to capture the data
once you know beat/downbeat what kind of structure can you put it into
it's not always clear how apt the MIR alorithms are by "accuracy" metrics
there are many applications we can think of that would be useful, but before that we need a representation of the data
Carlos "more than obvious" that generative models have many applications - e.g. sequencing and interpolating between "feels" and "styles"
if you have different generative models it's interesting from an explorative point of view - exploring space between styles
Georgio Mar - a drum machine helped him learn new ways of drumming
Analysis is "well defined" problem, but the creative side is not well designed
Objectively we should be able to find things that will help each community, but we need to be able to motivate reaserchers and convince them there is benefit
Composers are not ready to rencounce their freedom of creativity - from MIR you just want them to try it
How much MIR do you actually need? You can get it from musicological studies?
Yes, but you can use comparisons from MIR output against output from certain corpuses
How much corpus based stuff is at IRCAM?
What they want is other styles of improvisation using annotated corpus with many annotated features
indirect MIR with generative processes - show off some of the informal rules
crowd sourcing - give generic tools to musicians and ask them to tweak them to a specific style, those paramaters back and you have a model of a specific style and you don't have to do anything
maybe this is soemthing that can be done in a corpus based style
Workshops, labs
I proposed a workshop/forum on interdisciplinary on this issue - maybe starting with a forum
real-time MIR workshop idea from City University
DML project at City - providing APIs
Similarity workshop at Anja Volk's institution
MIR people need musicologists, but musicologist might not need MIR
Existing (attempted) solutions - case studies
Having APIs or something you can try out is appealing - City's DML project is aiming to do this
Put something out there - but what??
DJ tracking software has solved some of these files - e.g. ableton live has lots of stuff in real-time
albleton does a good job or not - is it general or is it specificly taylored to beated 4/4
that's part of the problem as what you do with ableton is represented by the output - we should be a ble to know more e.g. strucutre, metrical relationships, where are the deviations from relationships
Case studies:
"Orkide"(sp?) uses a huge corpus of intrumental techniques and it tells you the combination of things you need to create a target timbre
Antescorfo
MIREX UX Grand Challenge that asks people to put a GUI on their MIR algorithms is at least a start but hasn't been successful this year, hardly any submissions
Nick Collins asked composers to compose music to break MIR systems (in a controlled way) - it helps understand the limits of your systems in a sytematic way - we should do more of this!!
Freeound has been out with the idea that it will help everyone - but they get no feedback
Massive mismatch between downloaders and uploaders (50000 downloads/ a thousand uploads - 250000 sounds)