Moaning, screaming and titillating laughs are just some of the sounds of pornography that we know of. The sound of sex does not need the mind of a genius or the imagination of a normal person in order to know what a porn video sounds like. What if these sounds will be the determining factor for our computers to know the signal to filter porn?
At present, there exists several computer-monitoring software that analyze images to discern if photos have explicit contents or not. However, most of them struggle to distinguish indecent imagery from the contents with large flesh-colored regions, such as a person in bathing suit or a close-up face.
Analyzing the audio for sexual resonance could probably solve the problem most parents face with their kids. Electrical engineers MyungJong Kim and Hoirin Kim at the Korea Advanced Institute of Science and Technology in Daejeon, South Korea are putting together a signal-processing technique called the Radon transform, which creates spectrograms of a variety of audio clips.
From the engineers’ study, they found out that a person’s speech are normally low-pitched, while musical clips have a wide range of pitches – both vary gradually over time. On the contrary, sounds of pornography are usually higher-pitched and periodically repeating in a fast manner.
With these sound characteristics and principles in mind, the researchers gathered different audio clips taken from online videos. Using a statistical model to classify sounds as pornographic or non-pornographic, samples used included non-sexual audio clips such as music, movies and television shows.
Surprisingly, their model resulted to a 93% accuracy of identifying pornographic content from the set of test clips they have sampled.
The remainder of 7% had confusing sounds, such as background music, comedy laughter, loud audience cheers and dramatic cries. These were mistakenly tagged as sounds of pornography due to its similar spectral characteristics to sexual sounds.
While the research has proven to be working, there are still loopholes where pornographic contents can seep through. With image-content analyzers, software can do its job with just a single frame, whereas the audio method will require longer audio test clips in order to work.
Perhaps the combination of both the image and audio method will be far better in keeping the explicit contents away from children. As time passes by, kids using the computer and the Internet are getting younger by the day. With contents readily available online, their young minds easily get polluted with things that are meant for adults.
Let’s see where this development goes, as the researchers present their work at the International Workshop on Content-Based Multimedia Indexing in Madrid, Spain.
This article is my seventeenth contribution to Manila Bulletin -- one of Philippines' leading broadsheets -- published on May 30, 2011 (Monday) in the TechNews Section. You can view the PDF version here.
Photo Source: Google Images
0 comments:
Post a Comment
Don't choose Anonymous! Indicate your name or alias please.