Cypher has developed a technology that uses deep neural networking to remove background noise from voice communications.
Utah-based startup Cypher has developed a technology that uses deep neural networking to recognize the elements of speech to remove background noise from voice communications.
In addition to the neural network technology, the solution uses math and pattern recognition techniques to detect human voice, isolate the primary speaker’s voice and ignore all background noise, according to Cypher CEO John Walker and chief strategy officer John Yoon.
In recent testing of the technology's automated speech recognition filtering on Amazon Echo's Alexa, accuracy improved by as much as 121 percent when background noise was present, according to Walker. He added that the testing included rather complex queries like what the weather is like in a particular city, not just simplistic questions requiring one-word answers. In total, the test included 195 open-ended queries.
The ASR Filtering solution is entirely software-based and can be added on to existing chips, according to the company.
The issue of background noise has become more bothersome as consumers rely on their mobile devices for a majority of their communications, often in noisy environments, and increasingly rely on personal assistants such as Alexa, for a variety of functions, Walker says.
In a survey that Cypher conducted with Harris earlier this year, 61 percent of mobile phone owners said they had to end calls due to background noise. The noisiest environments are restaurants or cafes (67 percent), areas where babies are crying (49 percent), busy city sidewalks (47 percent), and airports (39 percent).
The survey also found that that nearly three quarters (74 percent) of mobile phone owners would be interested in a mobile phone function or feature that allows them to control whether a call recipient can hear the background noise.
The poll findings show the need for a better solution for dealing with background noise, according to Cypher executives.
"All of the other noise cancellation solutions have looked at it as a sound problem," Yoon says. "Our CEO, John Walker, and the bulk of our team looked at it as a computer science problem, so they looked at deep neural networks that instead of looking to suppress noise, looked to use a speech matching model."
So the technology finds the elements of speech and sends just that through the mobile device or through the personal assistant processing engine, according to Yoon. As a result, the Cypher technology performed as much as three to four times better than the best existing noise cancellation solution, Yoon said.
The technology will be piloted this fall with Cisco in the public safety market (police, fire, etc.), where background noise is the rule rather than the exception, Walker said.
Citing the survey, Walker added that phone carriers, consumers, and businesses providing personal assistant solutions and the people using them are all unsatisfied with the state of background noise cancellation solutions today, so he expects the Cypher technology to gain significant traction in the market within a year.