• by Laurie Sullivan  @lauriesullivan, October 11, 2016

Two particle physicists seeking dark matter have made it possible to search and identify sound clips in recorded audio, and are offering journalists the ability to use the tool for free until Election Day.  

Noah Shutty and Scott Stephenson got the idea for the startup after seeking a better way to search through their own audio files. One report suggests the need led Shutty to develop the neural net-based artificial intelligence engine known known as Deepgram.

The artificial intelligence engine makes searching for keywords in speeches, private conversations and customer service phone calls faster and less expensive, according to the company's Web site.

While the tech gives advertisers the ability to identify when someone mentions specific keywords in a customer support call or a brand name in a video clip posted to YouTube or another site, the company this week introduced the tool to journalists -- those who want to "find damning soundbites' in videos and recording of both presidential candidates.

The user must upload an audio file to the service, which uses Amazon Web Services. The file can contain anything from phone calls or videos, including a YouTube URL. The technology processes the speech. Rather than trying to translate sounds into words, the technology uses phonetics and allows individuals to search for words by the way they sound.

Deepgram launched out of the Y Combinator’s Winter 2016 program. The company says the technology can find results four out of five times, with an 80% accuracy rate, compared with 20% speech-to-text accuracy ratio. 

While the search engine can find words in the audio file, it also can analyze the inflections in the tone of words found in the recording, according to Levi King, who supports business development for Deepgram. He said the technology can identify inflections in the caller's voice to determine the possibility of a sale from a customer support call and use the audio to run A/B test on campaigns that include audio. It also can determine the possibility of conversion rates related to a customer service call.