| |
Abstract:
some spatial region of a fictitious "topology space" it is
possible to naturally define neighbouring states of any state as
those which are connected in that space. The transition matrix of
the HMM can then be constrained to allow transitions only between
neighbours; this means that all valid state sequences correspond to
connected paths in the topology space. This strong constraint makes
structure discovery in sequences easier. I show how such
constrained HMMs can learn to discover underlying structure in
complex sequences of high dimensional data, and apply them to the
problem of recovering mouth movements from acoustic observations in
continuous speech.
|
