SoFI Streaming Music using Song Form Intelligence

1
SoFI Streaming Music using Song Form
Intelligence

• By Jonathan Doherty
• Supervisors Dr. Kevin Curran, Prof. Paul Mc
  Kevitt
• School of Computing and Intelligent Systems
• Faculty of Engineering
• University of Ulster, Magee

2
Objectives of SoFI

• Pattern matching within songs
• Determine replacements based on pattern matches
• Client based error concealment of packet loss
• Possibly model song semantics

3
Previous Research

• Liang et al. (2001) sender based by adjusting
  packet size
• Ngo et al. (1999) sender and receiver based
  using a random order for sending packets
• Windows Media Player, Apples Quick Time
• Real Audio, Shoutcast

4
Song Form and Structure

• Western tonal music
• Verse, chorus, verse, chorus
• Music pattern matching
• SEMEX (Lemstrom and Perttu 2000)
• CAMUS 3D (Miranda 2001)
• Ghias et al. (1995) Query by humming

5
Project Proposal

• To match lost/late packets with previously
  received similar portions of songs
• Cognitive approach to error concealment
• MPEG-7 descriptors for tagging sections of songs

6
SoFI General Architecture
Pattern Matching and Replacement
Audio Player
Packet Loss Identifier
BUFFER
7
Comparison of Streaming Approaches
8
Software Analysis

• Programming Tools
• Java
• C
• VB .Net
• Streaming Applications
• Shoutcast
• Real Player
• Quick Time

9
Project Plan
10
Conclusion

• The objectives of SoFI meet challenging problems
  in streaming audio
• Maintain high quality audio streaming on bursty
  and bandwidth constrained networks
• Extend the QoS (Quality of Service) protocols to
  improve users listening experience
• Success of SoFI will be tested against accuracy
  in pattern matching and amount of packets
  replaced in relation to total number of packets
  lost