ZemPod: A semantic web
approach to podcasting
Journal Of Web Semantics 2008
Oscar Celma, Music Technology Group, Spain
Yves Raimond, Centre for Digital Music, UK
August 31th, 2009
Contents
2
Introduction
Background
System architecture
Usage scenario
Conclusions
Introduction [1/2]
Podcast
Portmanteau of the “iPod” and “broadcast”
A media file distributed in Internet
Use syndication feeds
Explosion in popularity of mobile devices
Make syndication model more attractive
Thousands of audio podcasts are available
on the net
3
Introduction [2/2]
There are some limitations of podcasting
No formal description
Only textual description available in HTML
No information about the contents of a podcast
session
Consists of a single audio file
Difficult to seek into one of the music tracks
To overcome these limitations
Using traditional audio signal processing
Speech/audio segmentation
Audio identification
4
Adding semantics to the podcast
Contents
Introduction
Background
Multimedia web syndication
Speech/music segmentation
Audio identification
The music ontology
System architecture
Usage scenario
Conclusions
5
Multimedia web syndication [1/2]
File format used for syndication
RSS
Really Simple Syndication (RSS 2.0)
Rich Site Summary (RSS 0.91 and 1.0)
RDF Site Summary (RSS 1.0)
Atom
To standardize feeds notation and autodiscovery
Due to some limitations and incompatibility versions of
the RSS family
6
Multimedia web syndication [2/2]
Example of RSS
7
Feeds and the semantic web
Atom/Owl
Aims at capturing
the semantics of the
Atom syndication
format
Feed
Attached metadata
Entry
Holds a text content
8
Speech/music segmentation
Discriminating between speech
(or spoken content) versus music
Achieving an automatically and meaningful
segmentation of a podcast session
Speech/music segmentation methods
9
Gaussian Mixture Models (GMM)
Support Vector Machines (SVM) classifiers
Combination of standard Hidden Markov
Models and Multilayer Perceptrons
Audio identification
Allows identification of unknown music
Audio fingerprint
A unique, compact code derived from
perceptually relevant aspects of a recording
Usages
Identification
Authentication
Content-based key generation
Content-based audio retrieval and processing
Hidden Markov Models (HMM)
10
Can precisely model temporal evolution of
audio signals
Music ontology [1/2]
Create a formal framework
Describing music-related information
Covering complex editorial information
External Ontologies used by Music Ontology
OWL-Time ontology
Describing the temporal content of Web
Interval, Instant
FRBR
Functional Requirements for Bibliographic Records
Work, Expression, Manifestation, Item
FOAF
11
Friend Of A Friend
Person, Group, Organization
Music ontology [2/2]
Describing a music production workflow
12
Contents
Introduction
Background
System architecture
RDFizing a podcast session
Access and workflow
Awareness of feeds
Resource identifiers
Usage scenario
Conclusions
13
System architecture
Main goal is
Analysing and
decomposing a
given podcast audio
file
RDFizing the
podcast information
14
The system
segments the
audio file into
speech and
music sections
15
Apply speech
recognition to
extract a list of
textual terms
16
Weight terms’
relevance
according to a
dictionary of
musical terms
17
Recognize
music chunks
using
fingerprinting
18
Query a
metadata
repository to
get basic
information
with the track
19
RDFizing a podcast session
To describe the semantics of a podcast
Using Atom-OWL and music ontology
“From 0 to 2 min, there is someone
speaking about Michel Jackson, then there
is a recording of a ‘Billie Jean’ in 1983”
Using 2 sub concept of the Event
MusicSegment
Temporal region holding music
SpeechSegment
20
Temporal region holding speech
Access and workflow
REST interface
Representational state transfer
Style of software architecture for distributed
hypermedia systems such as WWW
Allow us to access the podcast service
http://zempod.net/
Considering the podcast service is available
21
Access and workflow
- Awareness of feeds
POST
USER
201 (Created)
http://zempod.net/feed
Location Identifier
http://zempod.net/feed/4567
Internal representation of this feed
Music ontology/AtomOWL
Can be queried through SPARQL
22
Access and workflow
- Resource identifiers
MO/AtomOWL are designed as a
hierarchical URI space
Feed
Supports a syndication
http://zempod.net/feed/{FEEDID}
Entry
Holds a text content
http://zempod.net/feed/{FEEDID}/entry{ENTRYID}
Item
Actual contents
http://zempod.net/feed/{FEEDID}/entry{ENTRYID}/
item{ITEMID}
23
Contents
Introduction
Background
System architecture
Usage scenario
Submission of the original feed
Analysis of the new entries
Semantic description of the new entries
Conclusions
24
Submission of the original feed
Original feed
POST
http://www.ourmedia.org
/user/billy2rivers/mrss
201 (Created)
http://zempod.net/feed
Location Identifier
http://zempod.net/feed/1234
25
Analysis of the new entries
Processing a new podcast session
26
Semantic description of the new entries
USER
27
GET
http://zempod.net/feed/1234
Conclusions
To solve limitations of podcasting
No formal description of a podcast
Difficult to seek into one of the music tracks
Using traditional audio signal processing
Speech/music segmentation
Audio identification
Using semantic web techniques
Transform the current RSS to the Atom/OWL
28
It will ease some important music
information retrieval tasks
Related Ontology – MO/Event
To express the production process of a pi
ece of music
The main sub-classes of event
Performance, Recording, Arrangement,
Composition
29
Related Ontology - FRBR
Functional Requirements for Bibliographic Records
서지 레코드의 기능상 요건
목록규칙이나 목록의 완성을 의도하는 개체-관계 모델
서지 레코드의 구조와 관계
목록규칙 제정과 시스템 디자인을 위한 정확한 어휘 제공
30
WORK
EXPRESSION
MANIFESTATION
ITEM
저작
표현형
구현형
개별자료
is realized
is embodied
is exemplified
실현되다
구현되다
사례가 되다
is owned
is produced
is created
has a subject
소장되다
제작되다
창작되다
주제로 하다
FRBR – Entities and Relationships (1)
Entities and Primary Relationships
31
FRBR – Entities and Relationships (2)
Entities and “Responsibility” Relationships
32
FRBR – Entities and Relationships (3)
Entities and “Subject” Relationships
WORK
has as subject
has as subject
has as subject
33
WORK
EXPRESSION
PERSON
CONCEPT
OBJECT
MENIFESTATION
ITEM
CORPORATE BODY
EVENT
PLACE
MusicBrainz
34
© Copyright 2025 Paperzz