social network

Different artificial intellects could not decide what sounds on the record – “Yanni” or “Laurel”

It seems that the phenomenon leads to disagreements even among computers.

Sonix compared different speech recognition systems based on artificial intelligence using the phenomenon of “Yanni” or “Laurel.” Engineers missed the recording through the tools of Google, Amazon, IBM and through their own development.

Artificial intelligence Google and Sonix from the first time recognized the correct value – “Laurel”. However, the Amazon and IBM systems had difficulties: one heard the words “year old” (the old year), and the second recognized “Laurel”, but once again confused the word with “yeah role”.

Above - Amazon Transcription, at the bottom - IBM Watson
Above – Amazon Transcription, at the bottom – IBM Watson

In a conversation with TechCrunch, Sonix CEO noted that recognizing the human voice is a complex process, because there are many different variations of voice, rhythm, accent and frequency.

The reality is that different companies can optimize different templates used, so the results may differ.

Jamie Sutherland
CEO Sonix

The editor of the publication, David Coldewey, suggested that different results may arise because AIs have been trained at different frequencies of the human voice. Such an explanation seemed reasonable to the journalist.

On May 16, a record was distributed in social networks , on which the announcer pronounced only one word, but users heard different versions: some “Laurel”, and others – “Yanni”. The scientists decided that because of the frequency characteristics of the sound, people depending on technology and age hear it differently.

The next day, journalists Wired found out where the meme came from and what actually sounded on the record. It turned out that the phenomenon was discovered by a ninth-grader who did her homework when she met the incomprehensible word Laurel and decided to find it in the online dictionary

Back to top button