Music Information Retrieval in P2P Networks - Intelligent Music Information Systems: Tools and Methodologies - page 209

Figure 2. Feature extraction process

Figure 3. The ML algorithm

Query initialization. The querier assigns to R the entire query sequence (plus
the few query coefficients) and propagates (broadcasts) it to all its neighbors.
Reception of R. Upon the reception of R , each MH P probes its indexes,
resolves the false-positives, and produces a list of results (only true-positives).
The answer-set is propagated back to the querier, by broadcasting it to all the
neighbors of P (backward phase). Accordingly, should there be available h , R is
conveyed to all P's neighboring MHs (forward phase).
Reception of an answer-set. Each MH P , that is not the querier, receiving
an answer-set, continues the propagation (backward phase) to all its neighboring
MHs as long as there is available h .
An answer-set reaches the querier. When an answer-set reaches the querier,
then the results are immediately presented to the user.

to a query, the root is initially retrieved and its
entries that intersect the query are only further
examined recursively until reaching a leaf. All
nonintersecting nodes are not included in the
search. An example is given in Figure 2b. There-
fore, when searching for similar subsequence,
candidates from the R*-tree are first retrieved.
The candidates are ranked so as to process the
most promising ones first and then, those candi-
dates are examined against the provided query
representation. When the latter is reduced (as in
the case of transcoding that will be explained),
false-positives are still possible. Nevertheless,
their number is significantly reduced. More
details about indexing can be found in Karydis,
Nanopoulos, Papadopoulos, and Manolopoulos
(2005a).

searching algorithms

This section presents the two algorithms that
implement the searching procedure, as described
in Karydis et al. (2006). The first is based on
simple choices concerning the representation R
of the query sequence and its propagation dur-
ing the forward and backward phases, while the
second is based on more advanced choices with
respect to the latter issues.

Algorithm Based on Maximal Query
Representation

A simplistic approach for the representation R is to
set it identical to the query sequence. The advan-
tage is that no false-positives occur, since when a
possible match has been found by index probing,