Wyniki wyszukiwania - BazTech

1

Sound Isolation by Harmonic Peak Partition for Music Instrument Recognition

Zhang X., Raś Z.W.

Fundamenta Informaticae

|

2007

|

Vol. 78, nr 4

613-628

EN

Identification of music instruments in polyphonic sounds is difficult and challenging, especially where heterogeneous harmonic partials are overlapping with each other. This has stimulated the research on sound separation for content-based automatic music information retrieval. Numerous successful approaches on musical data feature extraction and selection have been proposed for instrument recognition in monophonic sounds. Unfortunately, none of those algorithms can be successfully applied to polyphonic sounds. Based on recent successful in sound classification of monophonic sounds and studies in speech recognition, Moving Picture Experts Group (MPEG) standardized a set of features of the digital audio content data for the purpose of interpretation of the information meaning. Most of them are in a form of large matrix or vector of large size, which are not suitable for traditional data mining algorithms; while other features in smaller size are not sufficient for instrument recognition in polyphonic sounds. Therefore, these acoustical features themselves alone cannot be successfully applied to classification of polyphonic sounds. However, these features contain critical information, which implies music instruments' signatures. We have proposed a novel music information retrieval system with MPEG-7-based descriptors and we built classifiers which can retrieve the important time-frequency timbre information and isolate sound sources in polyphonic musical objects, where two instruments are playing at the same time, by energy clustering between heterogeneous harmonic peaks.

2

Cooperative answering of queries based on hierarchical decision attributes

Raś Z. W., Dardzińska A., Zhang X.

Computer Assisted Mechanics and Engineering Sciences

|

2007

|

Vol. 14, No. 4

729-736

EN

This paper considers decision systems with decision attributes which are hierarchical. Atomic queries are built only from values of decision attributes. Queries are constructed from atomic queries the same way as we construct terms in logic using functors {+,*,Ź}. Negation symbol "Ź" is only used on the atomic level. Queries are approximated by terms built from values of classification attributes. We only consider rule-based classifiers as the approximation tool for queries. When a user query fails, then the cooperative module of the query answering system (QAS) constructs its smallest generalization which does not fail and which is approximated by rules of the highest confidence discovered by the classifier. Two interpretations of queries are proposed: user-based and system-based. They are used to introduce the precision and recall of QAS. The implementation of QAS follows system-based interpretation. Automatic indexing of music by instruments and their types is an example of the application area for the proposed approach.

3

Knowledge - discovery based flexible query answering systems

Raś Z. W.

Prace Instytutu Podstaw Informatyki Polskiej Akademii Nauk

|

2003

|

Nr 963

1-14

EN

Query answering systems are strictly dependent on models used to store the data. If these models are table-based and objects stored in them are described by weighted values of attributes, then query languages linked with these table-based models should be based on alphabets containing non-weighted values of these attributes and some logical connectives including: or, and, not. In this paper, we also give two examples of models used to store the data which are not table-based. The first one, introduced by Ras in [1] and called po-system, is hierarchical one with attributes and their values hidden. Ras shows in [1] how to reconstruct these hidden attributes and their values. The second one, called information tree, was introduced by Chen and Ras [16,17]. Since information trees are constructed from table based systems, attributes and their values are known a priori and they can be naturally used when constructing query languages for information trees. One way to make query answering systems flexible is to assume a hierarchical structure of their attributes [9,12,13,14,15]. The resulting query answering systems based on hierarchical attributes are often called cooperative [12,13,14,15]. One way to develop knowledge based query answering systems is to discover rules either locally or at remote sites (if system is distributed) and use these rules in a query answering process. There are two classical situations when it can be done. The first one is when attributes are incomplete so we may need rules to approximate the null values and the same way change the answer to a query. The second one is when users want to ask queries based on some attributes which are not listed in a local domain. Since these attributes are locally not available, we can only search for their definitions at remote sites and use them to approximate given queries [7,8,9,10]. This paper gives mainly an overview of the results presented in [1,5,6,7,8,9,10] with a goal to present new foundations for knowledge-discovery based query answering systems in a distributed scenario .

PL

Elastyczne systemy odpowiedzi na pytania oparte na wydobywaniu wiedzy Systemy odpowiedzi na pytania są ściśle uzależnione od modeli przechowywania danych. Jeśli modele te są tablicowe i dane przechowywane o obiektach są opisane w nich wartościami atrybutów z wagami, to języki zapytań budowane dla tych modeli powinny być oparte na alfabecie zawierającym te wartości atrybutów z pominięciem wag oraz spójniki logiczne, takie jak: lub, i, nie. W pracy tej również podane będą dwa modele przechowywania danych, które nie są tablicowe. Pierwszy z nich, zaproponowany przez Rasia w pracy [1] i nazwany po-systemem, jest modelem hierarchicznym, w którym atrybuty i ich wartości są ukryte. Raś w pracy [1] pokazuje, jak atrybuty te i ich wartości można odtworzyć. Drugi model, zaproponowany przez Chena i Rasia [16,17], nazwany został drzewem informacyjnym. Ponieważ drzewa informacyjne są konstruowane z systemów tablicowych, atrybuty i ich wartości są naturalnie znane i tym samym mogą być używane do konstrukcji języków zapytań dla drzew informacyjnych. Jednym ze sposobów budowy elastycznych systemów odpowiedzi na pytania jest założenie, że atrybuty systemu przechowującego dane są hierarchiczne [9,12,13,14,15]. Systemy odpowiedzi na pytania oparte na atrybutach hierarchicznych nazywane są często kooperacyjnymi [12,13,14,15]. Jednym ze sposobów budowania systemów odpowiedzi na pytania opartych na wiedzy jest wydobywanie reguł z systemów lokalnych lub odległych (jeśli system jest rozproszony) przechowujących dane i używanie tych reguł w procesie odpowiadania na pytania. Poniżej wymienione będą dwie klasyczne sytuacje, gdy wydobywanie reguł może być pomocne przy budowaniu lepszych systemów odpowiedzi na pytania. Pierwszy dotyczy sytuacji, gdy dane o obiektach są niekompletne i gdy reguły opisujące niekompletne atrybuty systemu przechowującego dane są użyte do zastąpienia tych niekompletnych danych wartościami przybliżonymi. Drugi dotyczy sytuacji, gdy użytkownik chce zadać pytanie używając atrybutów spoza dziedziny lokalnego systemu przechowującego dane. Atrybuty te są lokalnie niedostępne, tak więc, aby przetworzyć lokalnie pytanie użytkownika, ich definicje są wydobywane z systemów odległych i używane lokalnie do zastąpienia pytania użytkownika pytaniem przybliżonym [7,8,9,10]. Ta praca jest głównie przeglądem wyników opisanych w [1,5,6,7,8,9,10] i ma na celu przedstawienie nowych podstaw do budowania elastycznych systemów odpowiedzi na pytania opartych na wydobywaniu wiedzy w środowisku rozproszonym. Słowa kluczowe: systemy informacyjne, elastyczne systemy odpowiedzi, kooperacyjne i kolaboracyjne systemy, wydobywanie wiedzy, ontologie, zbiory przybliżone.

4

Global Action Rules in Distributed Knowledge Systems

Raś Z.W., Gupta S.

Fundamenta Informaticae

|

2002

|

Vol. 51, nr 1,2

175-184

EN

In papers [4,5], query answering system based on distributed knowledge mining was introduced and investigated. In paper by Ras and Wieczorkowska [3], the notion of an action rule was introduced and for its application domain e-business was taken. In this paper, we generalize the notion of action rules in a similar way to handling global queries in [4,5]. Mainly, when values of attributes for a given customer, used in action rules, can not be easily changed by business user, definitions of these attributes are extracted from other sites of a distributed knowledge system. To be more precise, attributes at every site of a distributed knowledge system are divided into two sets: stable and flexible. Values of flexible attributes, for a given consumer, sometime can be changed and this change can be influenced and controlled by a business user. However, some of these changes (for instance to the attribute ``profit'') can not be done directly to a chosen attribute. In this case, definitions of such an attribute in terms of other attributes have to be learned. These new definitions are used to construct action rules showing what changes in values of flexible attributes, for a given consumer, are needed in order to re-classify this consumer the way business user wants. But, business user may be either unable or unwilling to proceed with actions leading to such changes. In all such cases we may search for definitions of these flexible attributes looking at either local or remote sites for help.