Kedis ku xidhan hagaha · Dhexdhexaad Heerka

Kediska Ogaanshaha Deepfake Audio

Dib u eeg xaddidaadaha ogaanshaha maqal-qaboojinta, guud-maraynta cabbirka, saamaynta isku-buufinta, iyo waxa caddaynta la saxeexay ay samayn karto iyo ma dhisi karto.

Wadooyinka haga ee la xidhiidhaOgaanshaha Deepfake Audio
Su'aal 1 ee 8

Waa maxay hadafka udub-dhexaadka ah ee maqal-sheegaha been-abuurka?

Horudhac su'aal kasta8 su'aalo

Jawaabaha iyo sharraxaadaha waxay soo baxayaan markaad qaadanayso kediska.

  1. Waa maxay hadafka udub-dhexaadka ah ee maqal-sheegaha been-abuurka?
    • Si loo go'aamiyo in hadalku yahay mid dhab ah ama AI-synthesized
    • Si loo hagaajiyo tayada codka duubista
    • Si aad hadalka ugu qorto qoraal
    • Si loo cadaadiyo faylasha maqalka ah
  2. In “Does Audio Deepfake Detection Generalize?”, which feature comparison improved performance in the authors’ controlled evaluation?
    • Raw transcript tokens instead of audio
    • CQT or log-spectrogram features instead of mel-spectrogram features
    • A larger waveform file instead of a smaller one
    • Video frames instead of speech features
  3. Waa maxay sababta ogaanshaha qoto dheer ee maqalka loogu tilmaamay 'tartanka hubka'?
    • Baarayaasha waxaa lagu maamulaa saanad milatari
    • Faylasha maqalku way sii weynaanayaan
    • Hababka cusub ee isku dhafka ah ayaa ka baxsanaya qalabka lagu baaro been-abuurka hore
    • Waxay u baahan tahay laba qof si ay u shaqeeyaan
  4. Muxuu qiimaynayaa heerka ASVspoof ee caanka ah?
    • tayada ku talinta muusiga
    • Saxnaanta tarjumaada qoraalka
    • Xaqiiqda jiilka sawirka
    • Ka-hortagga been-abuurka/ ogaanshaha been-abuurka ee hadalka
  5. What can a valid C2PA Content Credential help a verifier check?
    • A signed record of an asset’s provenance and edit history
    • That every spoken claim in the audio is true
    • The speaker’s real-world identity in all cases
    • That a recording has never been compressed
  6. What input type did the ASVspoof 2021 Deepfake task target?
    • Uncompressed music recordings from a studio
    • Manipulated, compressed speech posted online
    • Images altered by face-swapping systems
    • Only live telephone conversations with known speakers
  7. How large was the real-world dataset collected in the cited audio deepfake generalization study?
    • 3.79 hours total, all authentic speech
    • 37.9 hours of deepfakes with no authentic audio
    • 37.9 hours total, including 17.2 hours of deepfakes
    • 172 hours total, including 17.2 hours of deepfakes
  8. What does the absence of a C2PA credential establish?
    • That the recording is necessarily a deepfake
    • That the recording has no edits
    • That the speaker is verified
    • Nothing by itself about whether the media is authentic or synthetic