New computer filter could detect porn content through sound

Scientists have developed a new weapon to fight against lewd footage appearing on computers through a 'porn filter', which recognises the sound of sex.
 
It's a large step-up from existing image-analysis systems, which struggle to distinguish between indecent imagery and innocent pictures with large flesh-coloured regions, such as a close-up face, reports New Scientist.

Electrical engineers MyungJong Kim and Hoirin Kim at the Korea Advanced Institute of Science and Technology in Daejeon, South Korea analysed a variety of audio clips for a "sexual scream or moan" while developing their technology.

They found that speech signals are normally low-pitched and musical clips have a wide range of pitches; both vary only gradually over time, pornographic sounds were higher-pitched and repetitive.

In tests, the model outperformed other audio-based techniques and correctly identified 93 per cent of the pornographic content.

The seven percent of footage it missed included confusable sounds, such as background music, or loud audience cheers.

Richard Harvey, a computer scientist at the University of East Anglia, described the technology as 'quite ingenious', but suggested it would be best to combine the audio filter with existing image-based ones.

He said: "Think of that scene in When Harry Met Sally (in which Meg Ryan fakes an orgasm while sitting in a diner).

"The audio is very clearly pointing in one direction, but the video is not," he added.