Researchers create porn audio detection system

first_imgThere are many cases where filtering through content for pornographic material not only makes sense, but is required. A great example of that is YouTube as it needs to keep its service free of porn, and anything classed as remotely adult sits behind an age warning.The main way of detecting porn is to use image filtering techniques. Such filtering has progressively got better. Detecting the difference between porn and an innocent scene that has a lot of skin on show will work 90+ times out of a 100. But such systems can always be improved, and researchers at the Korea Advanced Institute of Sciences and Technology believe they have a way to increase detection rates and lower false positives.Rather than looking at the video imagery, researchers Hoirin Kim and MungJon Kim focused on the audio instead. They discovered that the audio in pornography is quite distinct and therefore is easy to detect. They created spectograms of audio clips using a radon transform to represent the voice, music, and sound effects along a parallel path.The results showed a clear distinction between porn audio and that found in most movies, TV, and music. Porn is decidedly high pitched, quick changing and repetitive.Using what they had learned, the two Kims were able to classify different types of audio and distinguish them as porn sounds or not. When tested across multiple soundtracks, they achieved a 93% success rate.On its own, that’s a very reliable system, but if you combine it with image filtering too it takes the success rate up significantly higher and therefore will be desirable for anyone having to carry out automated porn searches.Read more at New Scientistlast_img read more