1 / 25

Tussentijdse presentatie

Perceptual declipping of audio signals through compressed sensing: algorithm design and evaluation. Tussentijdse presentatie. Naim Mansour. Promotor: Prof. dr. ir. Marc Moonen Assistent: Ir. Bruno Defraene. Overzicht. Onderwerp & doelstellingen (vermelding Steven) – 3 min.

lucio
Download Presentation

Tussentijdse presentatie

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Perceptual declipping of audio signals through compressed sensing: algorithm design and evaluation Tussentijdse presentatie NaimMansour Promotor: Prof. dr. ir. Marc Moonen Assistent: Ir. Bruno Defraene

  2. Overzicht • Onderwerp & doelstellingen (vermelding Steven) – 3 min. • Compressedsensing – 5 min. • Wat? • Theoretisch • Declipping (don’tforget perfect reconstruction) • CS & Declipping – 4 min. • Specifieke theorie • Eerder werk (INRIA, AxBe) • Kort: perceptuele component • Toelichting gemaakte keuzes & motivatie (2 keuzes) – 3 min. • Don’tforget frame length (basicallyall details) • Implementatie & resultaten (demo) – 5 min. • Planning, en plannen voor fase 2 – 2 min. • Dank & vragen

  3. Overview • Subject • CompressedSensing • CS & Declipping • Perceptual components • Extra: IRL1 • Implementation • Evaluation

  4. Subject • Declipping of audio signals • Through compressedsensing • Perceptual • Algorithm design & evaluation

  5. CompressedSensing: general • Candès, Romberg, Tao – 2006 • Recoversparsesignalfrom sub-Nyquistratesampledmeasurements • Consider the signals, sparse in a fixed basis : • Measurement basis selects reliable values fromsaccordingto ( is known as the sensing base): • Reconstructionthroughconstrained L0/L1 minimization:

  6. CompressedSensing: Choice of Lp • Solution equalstranslation of null(A)-planeby vector z • L0 & L1 lead tosparse solutions, L2doesn’t • L1 minimization is convex -> convex optimization, • L0 minimization non-convex -> greedyopt.

  7. CompressedSensing: AxBe model • Other possiblemodel (Bölcskei & Studer, – 2011) • In case of clipping, we considertobemeasurementincludingclipped samples (). No explicit measurement matrices, and, toobtainsparse error base (). • Recovery throughprojected Lpminimization:

  8. CompressedSensing: Recovery • Certain theoreticalboundsforperfect recovery of signal • Classical model (no noiseassumption): • AxBe model: • Coherence of a basis: measureof decorrelationin analysis domain • Fourierbase: DCT base: ,

  9. CS & Declipping: recovery • Recovery ability dependent on coherence of sensing base • Classical CS: Usage of pseudorandom measurement matrices (e.g. iid Gaussian sampling) leads to very low coherence • Declipping: reliable, “sampled” values in signal are unclipped ones-> clearly not pseudorandom! • Coherence of combined Fourier/DCT base with clipping sensing base = coherence Fourier/DCT • Recovery guarantees for DCT base ( reliable samples): • Perfect recovery of real audio signals practically always impossible, since

  10. CS & Declipping: • Missing samples will always lie beyond the clipping threshold • Lp minimization can be improved through introduction of additional linear constraints

  11. CS & Declipping: previouswork • INRIA • Bölcskei

  12. Perceptual components • Perceptualweighting matrix based on acousticloudnessperception • Psychoacousticallyoptimized (adaptive) basis

  13. Extra: IRL1 • Iterativelyreweighted L1 minimization (Candès, Wakin, Boyd – 2007)

  14. Implementation: general • 2 mainchoices • PCS throughbounded L1 minimization, usingperceptualweighting, Axy & AxBe models (furtherimprovementthrough IRL1) • PCS throughbounded L0 minimization, usingpsychoacousticwavelet basis, Axy & AxBe models • Incremental design: implementation & evaluationwith & without bounds, with & without perceptual components,…

  15. Implementation: Clipping

  16. Evaluation: general

  17. Evaluation: SNR vs. PEAQ

  18. SNR no guaranteefor audio quality!!

  19. Planning & future prospects

  20. Planning & future prospects

  21. Planning & future prospects • Semester 2 • Executepsychoacoustic experiments • Finish algorithms • Write finaltexts

  22. References • http://people.ee.duke.edu/~willett/SSP//Tutorials/ssp07-cs-tutorial.pdf • Recovery of SparselyCorruptedSignals blablabla

  23. ?

  24. Zalig Kerstfeest!

More Related