CallMiner Product Innovation Series: Q4 2024
CallMiner's, Bruce McMahon, shares key product updates from Q4 2024. New AI capabilities, including CallMiner AI Assist, add to the long list of AI ad...
The CallMiner team has been watching the Yanny-Laurel debate with great excitement. It is not every day that speech captures the world’s attention! After tirelessly polling our friends and family about what they hear, we decided: why ask your friends when you can ask a speech recognizer? Without further ado, here is what Eureka thinks:
If you extract the audio from the reddit video, it can clearly hear “laurel” or something similar at 16 khz and something like “yanny” at 8 khz:
16khz model: “laurel ”
8 khz model: “yeah me ”
However, the clip from vocabulary.com (allegedly the original source) results in a unanimous “laurel” for both:
16khz model: “laurel ”
8 khz model: “laurel ”
Based on the transcription results, we think that the reddit file was manipulated to include “yanny” in the lower frequency range. In the 16khz file, the higher “laurel” frequencies caused the recognizer to hear the “l”s and the “o/au”s; in the 8khz files the high frequencies were eliminated, leaving the “y”, “a”, “m/n”, and “e.”
The yanny-laurel spectrogram at 16khzThe yanny-laurel spectrogram at 8khz
Our brains, not unlike the speech recognizer, fill in the blanks depending on what we assume we’re about to hear. But I like to think that we are still a bit more flexible than the machines.
For a psycholinguistics and phonetics point of view, check out Suzy Styles’s Twitter posts and Benjamin Munson’s analysis.
So whether you hear Laurel or Yanny it’s an interesting debate and gets the rest of the world talking about speech analytics!
What do you hear? Tweet at us @CallMiner or post on our user community Engagement Optimization.
