[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Design and language for CASA implementation

Dear Andrew,

You wrote:
>I am currently in the process of designing an architecture for a CASA
>system for automatic transcription of music. I know that many people

I recommend dynamic Bayesian networks (DBN) for this purpose.
See for example (available online)
  Dynamic Bayesian Networks (Draft)
  To appear in Probabilistic Graphical Models, Michael Jordan.
  Kevin Murphy. November 2002.

Maybe you would not call DBNs a software "architecture",
but for me at least, it has provided exactly the kind of framework
that is needed for integrating knowledge from different sources
(acoustic data, internal models, musicological models etc.).
Probability theory is a solid "common ground" for integrating knowledge
and has served excellently in speech recognition.

I have designed one blackboard architecture in
  Klapuri. "Means of integrating audio content analysis algorithms".
  110th Audio Engineering Society Convention, Amsterdam, Netherlands, 2001.
  (available http://www.cs.tut.fi/~klap/iiro/index.html)
but personally, I nowadays ***discourage*** using blackboard
architectures. Matters of implementation easily
distract from thinking about the model-level framework.
Especially many young researchers (me included) have simply wasted time
with artificial-intelligence architectures (implementation) before
thinking about the computational theory and the algorithm for
attaining the desired analysis.

There is one approach to automatic transcription of music
which is still largely unexplored. That is the redundant
approach (see e.g. Bregman in "Computational auditory scene analysis,"
Rosenthal, Okuno (Eds)), where ***a lot*** of simple algorithms
that try to attain a same goal are put together. I am meaning
tens of bottom-up and top-down heuristics.
In this case (and maybe only here), the architecture plays a crucial role.

Anssi Klapuri    klap@cs.tut.fi   http://www.cs.tut.fi/~klap
Rongankatu 11 C 61, 33100 Tampere, Finland
Tampere University of Tech., P.O.Box 553, FIN-33101 Tampere, Finland
+358-3-3115 2124/-4954(work,fax) +358-3-2145252(home) +358-50-3648208(gsm)