KNOWN-AUDIO DETECTIONUSINGWAVEPRINT: SPECTROGRAM FINGERPRINTINGBY WAVELET HASHING

Inthis paper, wepresent anovel system fordetecting knownaudio. We start withWaveprint, anaudio identification system that, given aprobe snippet, efficiently provides reliable forced-choice ranking ofentries fromanaudio database. Foropen-set detection, wecan re-examine thebest-ranked matches fromWaveprint using simple temporal-ordering-based processing. Theresulting system has excellent detection capabilities forsmall snippets ofaudio that havebeendegraded inavariety ofmanners, including competing noise, poorrecording quality, andcell-phone playback. The system ismoreaccurate thantheprevious state-of-the-art system whilebeing moreefficient andflexible inmemoryusageand computation.