A recent study reveals that humans are only able to detect deepfake speech only 73% of the time. A recent study conducted by University College London (UCL) reveals that humans are only able to detect deepfake speech only 73% of the time, regardless of whether it's in English or Mandarin.