SIGNALS

It was an internet sensation, that's for sure. And a real head-scratcher, too. People listened to the same audio clip, and heard two completely different words. Much like the black/blue-white/gold dress craze that visually swept the internet three years ago, yanny/laurel did the same thing, but aurally.

Last week, a short audio clip went viral. A robot-tinged voice spoke a single word, but listeners disagreed on what the word was. When presented with the choice, some people heard "yanny" while others heard "laurel." Now, it's easy to misunderstand words. If I say I've got an appointment at the "harbor," you might misunderstand me and think I am getting a haircut at the "barber." But "yanny" and "laurel" are very dissimilar sounding; how on earth could people confuse them?

The reason is both simple, and subtle. But before we delve into it, if you haven't heard the file, listen to it here. In fact, even if you've already heard it, listen again, because you might hear it differently than before.

So, what's going on? For starters, as you might expect, in effect, both words are being spoken at the same time. More precisely, the audio file has energy distributed at frequencies that correspond to the vocal components of both words. There is no aural hallucination, illusion, or even a trick; the aural information for both words is in the file, thus creating ambiguity in intelligibility. People in the same room, listening to the same playback, would hear two different words.

Using this tool at the New York Times, you can play with the frequency response of the clip. In particular, you can sweep sweep a bandpass filter across the pertinent frequency range, and emphasize the lower-frequency laurel region or the higher-frequency yanny region. When the filter is somewhere in between, balanced between those regions, the choice is more difficult. There are other vocal components in both words that are not exclusively high or low frequency, but the high/low frequency switch seems to most clearly distinguish them.

Given this jumble of audio energy, several factors help explain why people hear different words. One possible factor is the playback device. If the frequency response isn't flat, certain frequencies will be emphasized and made more prominent. Another factor is the hearing response of the listener; again, different frequencies will be emphasized. It also depends on the playback volume; our ear's response curve varies greatly with volume level, gradually becoming flatter at louder listening levels.

Dr. Chris Bennett, an audio research professor at the University of Miami, worked up a detailed acoustic analysis of the file for me, explaining empirically why its content creates ambiguity. He also pointed out that the biggest factor may be what lies between the ears — the brain. We are very selective about what we hear and effectively "tune in" to certain parts of any audio signal. We are also very prone to suggestion; when someone tells us the bass is too loud, we are all too willing to suddenly begin to perceive loud bass.

If we could wipe the internet clean and start over, we could play the file, and ask what word people are hearing. From the jumbled contents of the file, we might get all sorts of heard responses. If we simply told people the word was "laurel" most people would hear that. In this case, we are being asked to choose between two words; this forces us into a necessarily ambiguous choice and not surprisingly, we get disagreement between both words. In the end, we hear what we want to hear.

The reality is that because of that whole brain thing, our senses are far from trustworthy. Jurors have convicted many a man based on "eye-witness" testimony but in fact our eyes routinely deceive us. I wonder how many people have gone to prison or worse because an eye witness positively identified a blue shirt, when in fact the shirt wasn't blue at all. Likewise our ears are no better. In this case, when confronted with an audio signal that offers a choice of recognizable words, it is no surprise that our brains arbitrarily pick one, or just can't decide at all.

Finally, for the record, it's "yanny." If you hear "laurel," you're just wrong.