Humans are actually quite good at having a discussion with a single person while blocking out other voices.
Surprisingly enough, though, technology has just lately been able to mimic this skill.
And that counts when it comes to the admissibility of audio evidence in court. Background noise can make it difficult to distinguish between speakers and what they are saying, which could render recordings unusable.
Wave Sciences’ creator and chief technical officer, Keith McElveen, is an electrical engineer who first became interested in the issue while working on a war crimes case for the US government.
Our goal was to identify the person who gave the order to massacre civilians. I discovered the nature of the “cocktail party problem” when I listened to recordings of multiple voices speaking simultaneously that were among the evidence, the man claims.
“I had been successful in eliminating background noise from speech, such as air conditioners, fans, and car sounds, but when I tried to eliminate speech from speech, it turned out to be one of the classic hard problems in acoustics as well as a highly challenging problem.
