The fight against Internet smut is an old one, and a challenge ofsysiphusian futility. But researchers MyungJong Kim and Hoirin Kim at the Korea Advanced Institute of Science and Technology have presented a new weapon in the fight for decency. They’ve recently unveiled new software that analyzes the sound of video clips to determine whether or not the video is pornographic. Like Justice Potter Stewart, this software should know it when it it hears it.
The new technique uses the Radon transform (you know, “the integral transform consisting of the integral of a function over straight lines” named after Johann Radon) to analyze the audio of suspect materials, and determine if the contents fit the profile of pornography. After analyzing numerous half-second audio clips with their system, the researchers released several shocking conclusions about the sounds of sex. From the New Scientist:
[S]peech signals are normally low-pitched and musical clips have a wide range of pitches; both vary only gradually over time. In contrast, pornographic sounds tend to be higher-pitched, change quickly and also periodically repeat.
When tested against other audio screening software, this model performed quite admirably with a 93% success rate. Of course, the system is not perfect and was inconclusive about clips that featured background music. Additionally, it struggled to discern some clips and registered them as false-positives. Again, from the New Scientist:
Comedy shows with laughter were also sometimes mistaken for pornography, as the loud audience cheers and cries share similar spectral characteristics to sexual sounds.
“Loud audience cheers?” Sounds like someone is having a way better time than me. That aside, were this technology to be combined with existing image analysis software, it could be a powerful new tool against eSmut. If you’re into that kind of thing.
(via New Scientist)