Study: Audio Processing Makes Real Speech Look Synthetic to Detectors
A researcher tested 13 synthetic-speech detectors on 47 real recordings and found normal audio processing (codec, denoise, EQ) makes them flag genuine speech as generated. A codec pass shifted 9 of 13 detectors, and 1984's Griffin-Lim shifted 12 of 13.
- Codec pass shifted 9 of 13 detectors, 3 on all 47 recordings
- Griffin-Lim without neural networks shifted 12 of 13 detectors
- Denoise and mastering shifted two detectors by -2.95 and -4.22
- Detectors from two families fail on opposite processing types
Read next
AI