We've long been impressed with the Moto X's touchless control feature, which fires up Google Voice assistant whenever you say "Ok Google," even if the phone is asleep. However, Motorola's technology is proprietary and limited to just that one phone. Fortunately, Audience, a leading manufacturer of audio processing chips for mobile devices, has an always-on voice recognition solution that any OEM can purchase. Announced today, the company's VoiceQ technology detects human speech and then wakes the phone if and only if you say the correct activation phrase.
A feature of Audience's new eS700 series of audio processing chips, VoiceQ remains in a very low power state while the phone is idle, using just a few milliwatts to power one of the device's microphones while it listens for the sound of a human voice. The chip is smart enough to know the difference between a human voice and other types of sound so it will not take further action unless it hears a person speaking. Once it determines that a sound is speech, it goes into a slightly higher power state where it checks the speech against the magic word, which could be "Ok, Google" or any other phrase that the user or software has designated. If VoiceQ detects that you've spoken the activation phrase, it then brings the phone to a full power state and immediately beings processing any commands you've issued.
Maintaining a low-power state while listening for the activation phrase is critical for any always-on voice technology. If your phone had to remain awake just to listen for your voice or if it had to wake up just because it detected a sound and had to check to see if it was a voice saying the correct words, it would run out of juice very quickly. Most phones have 2 to 3 microphones so they can better separate background noise from your voice and even keeping all of those powered at all times would use up more battery than VoiceQ's single-mic listening state.
We had a chance to test out Audience's VoiceQ technology at CES 2014 and we were impressed with how accurate and power-efficient it was. For demo purposes, the company had a smartphone attached to a large screen power meter. When we spoke the activation phrase "Ok earSmart," the meter awoke and displayed a power trend line showing how the phone was just consuming a couple of milliwatts of power when idle, which then increased to about 6 milliwatts while VoiceQ checked our speech input against the activation phrase. Once it determined that we had said the phrase, the phone woke completely and consumed maximum power.
In addition to VoiceQ functionality, Audience's eS700 audio processing chips also provide a new level of noise reduction, correctly identifying and filtering out wind noise, crowd noise and other irritants that interfere with audio processing and call quality. Speech restoration technology tries to fill in missing syllables when someone's voice is cut off.
Audience is sampling its eS700 chips to device vendors right now. We expect to see a number of phones and tablets with VoiceQ later in the year.