You may not have heard of Audience, but odds are one of its chips is inside your smartphone. The company's audio processors can be found in over 160 mobile devices, including multiple generations of iPhones and all the latest flagship handsets. Audience silicon is also used in Google's Nexus 10 tablet and Dell's Vostro 5460 ultrabook. Now, the firm has a new family of products designed with always-on voice recognition in mind.
Unlike Audience's previous chips, which have strictly been audio processors, the latest eS700 Series adds codec duties to the mix. The integrated codec is situated on the same chip package, but it's a separate piece of silicon—and an optional one at that. Versions of the eS700 will be available with and without the codec onboard.
Audience claims the eS700's integrated codec has "best-in-class" specifications, but the real magic happens in the audio processor. That's where we find VoiceQ, an always-on voice recognition function tuned for maximum power efficiency. VoiceQ's lowest-power state listens for human voices continuously. When one is detected, VoiceQ shifts into a keyword-spotting state that listens for a specific phrase, such as "Ok Google." If the correct key phrase is spoken, VoiceQ start buffering incoming audio and signals the host system to wake up. The buffered audio is then passed along to the host in a continuous stream.
The first two stages consume relatively little power compared to waking a typical handset, so accurate phrase detection is key to preserving battery life. Audience says a single false positive draws as much juice as two hours of the keyword-listening mode.
VoiceQ supports custom pass phrases, and it's supposed to be smart enough to recognize different voices. It's also joined by a host of additional features, including a fourth-generation processing engine that suppresses wind noise and improves clarity for awkward microphone angles. There's also a 360° speakerphone mode designed for group calls and in-car use.
The eS700 doesn't perform speech recognition, but it does massage voice signals to improve the accuracy of third-party translation software. This "assist" functionality tweaks the audio for machine rather than human listening, and it's compatible with all the major speech recognition engines. The latest incarnation is supposed to improve accuracy in noisier environments, which can be particularly challenging for accurate speech recognition.
Audience says the first devices to tap the eS700 Series will arrive in the first half of the year. I suspect smartphones will come first, but the chip should work its way into tablets, and I wouldn't be surprised to see it pop up in convertible notebooks.