Se ha denunciado esta presentación.
Utilizamos tu perfil de LinkedIn y tus datos de actividad para personalizar los anuncios y mostrarte publicidad más relevante. Puedes cambiar tus preferencias de publicidad en cualquier momento.

Libros relacionados

Gratis con una prueba de 30 días de Scribd

Ver todo

Audiolibros relacionados

Gratis con una prueba de 30 días de Scribd

Ver todo

Software Dev

  1. 1. Keith Lohnes Pat Ruppel Adrian Wisernig
  2. 2. "  Determine best language   Speed   Robustness   Portability   Available Libraries   Ease of Implementation  Experiment with different algorithms Proto-Prototypes
  3. 3. "  Java  Tritonus  jLayer  jTransforms Scratch
  4. 4. "   Java   Library for extracting features from audio files   28 implemented features   GUI for testing   XML driven   Wide Array of formats supported jAudio
  5. 5. "  C/C++  Multimedia Fingerprinting Library  Supports Video, Images & Audio  GPLv3 pHash
  6. 6. " Project Requirements pHash Audio Fingerprinting Video,Image & Audio Fingerprinting Substring Comparison Only a few seconds are needed for match Speed C/C++ Runs on Lab Computers( Ubuntu) Open Source/Linux friendly Confidence Level Bit Error Rate between hashes( 0-1) Comparison
  7. 7. "  Compiling code on Debian/Ubuntu  Portability (sudo, etc)  Library dependencies  Developed mainly to prove thesis  Lack of documentation  Lack of widespread use (apt-cache rdepends libphash0)  Lack of C++ knowledge There is no Silver Bullet
  8. 8. "  IRC  Stack Overflow  Mailing list   Discovered several bugs associated with compilation   Will be part of next version of pHash But with a little help…
  9. 9. "   Bash Script   C++ File (87 lines) Project Overview
  10. 10. " Bash Script
  11. 11. " C++
  12. 12. " Calculate Hashes
  13. 13. " Determine Match
  14. 14. " Done
  15. 15. "  Bark Scale - Looks only at frequencies that humans hear  Frame size - 0.21 seconds  The value is based on the bark scale values of the frame FFT spectrum  Hash computed from temporal and spectral differences on the bark scale.  Compares Bit Error Rate between different hashes, normalized for time. Yeah but…
  16. 16. "  http://www.phash.org/docs/design.html  http://www.phash.org/docs/pubs/ thesis_zauner.pdf Learn More
  17. 17. "   Use threads to generate hashes concurrently   Make format detection more robust   Better way of determining % of confidence  Refactor Improvements
  18. 18. " End ?

×