Mimicking the Human Brain Could Give Us More "Human-Like" Voice Assistants, Researchers Say
By processing speech in a way that mimics the human brain, a pair of researchers believe they've come across a superior approach.
A pair of researchers at the Japan Advanced Institute of Science and Technology (JAIST) have come up with a way of making virtual assistants more human-like — by mimicking how the human brain works with speech.
"In humans, the auditory periphery converts the information contained in input speech signals into neural activity patterns (NAPs) that the brain can identify. To emulate this function, we used a 'matching pursuit algorithm' to obtain 'sparse representations' of speech signals, or signal representations with the minimum possible significant coefficients," explains Masashi Unoki, professor at JAIST and senior author of the paper.
"We then used psychoacoustic principles, such as the equivalent rectangular bandwidth scale, gammachirp function, and masking effects to ensure that the auditory sparse representations are similar to that of the NAPs."
In effect, the pair's approach ditches the normal method of running a voice-responsive virtual assistant system — where speech is converted to text then processed for recognized instructions and commands — in favor of something closer to the way humans perceive speech.
To test the system, Unoki and first author Dung Kim Tran opted to analyze 630 speech samples from different speakers then re-synthesize them before measuring the result for sound quality. The output of the system, they found, was comparable to the original input signals — a finding backed up by a practical experiment in pattern matching with speech from a set of six speakers.
"Our results showed that the auditory sparse representations produced by our method can achieve high quality re-synthesized signals with only 1066 coefficients per second," Unoki says. "Furthermore, the proposed method also provides the highest matching accuracy in a pattern matching experiment."
The pair's work has been published under open-access terms in the journal IEEE Access.