New Cirrus Logic Processors Geared Toward Voice Accuracy

Cirrus Logic began shipping voice processors with its next-generation SoundClear technology that are designed to boost voice recognition accuracy in smartphones, tablets and wearable electronics by eliminating background noise and echo, Carl Alberty, marketing director, told Consumer Electronics Daily. The technology could be in handsets for the holiday season, he said.

A consumer’s relationship with a smartphone “should be positive,” Alberty said, saying voice interactions with smartphones today “leave a bit to be desired.” Cirrus’s goal is to enable consumers to have an intelligible conversation in “any environment” such as a quiet room, loud bar, train station or car and not have background noise transfer to the person at the other end of the conversation. Another goal is to improve speakerphone experiences by canceling out the echo “that can be introduced very easily with speaker and microphone being so close to one another,” he said.

The new Cirrus Logic CS48LV12/13 processors enable use cases such as multi-person speakerphone calls, and the CS48LV13 adds ASR Enhance, a dual-microphone noise reduction technology said to improve the reliability and accuracy of automatic speech recognition engines such as Apple’s Siri and Google Now when operating in noisy environments.

The CS48LV13 also supports what Cirrus Logic calls the industry’s “lowest power implementation” of TrulyHandsfree technology from Sensory that enables users to control a phone by voice. Some handsets do that, but Alberty said early designs “didn’t really get the solution quite right” because they weren’t always on, meaning they weren’t always in listening mode for a voice command. “You had to fire up an application for that technology to work,” which he said made the voice technology more of a “novelty” than a compelling use.

Motorola’s Moto X, launched last fall, was the first smartphone to offer a “truly embedded, always-on capability” for voice wake up, Alberty said. He gave the example of someone reading a recipe from a smartphone and not wanting to have to touch the phone’s screen to wake up the device and find the next recipe step. Device makers have held back from including the feature because it requires a phone to be in an always-on mode which includes “having a power profile that allows it to be always on,” he said. The implementation made possible by the CS48LV13 is “significantly lower power” than that of the Motorola phone currently in the market, which Cirrus does not provide, he said. In its solution, Cirrus combines TrulyHandsfree from Sensory Inc. with its own SoundClear ASR Enhance pre-processor, “ultra-low” power voice detection, optimized hardware and intelligent system control to enable a device to be active and on, he said.

Cirrus’s goal is to “push power down so much” that it can introduce new signal processing solutions for problems that it deems necessary to provide quality noise reduction, Alberty said. Many handset makers already mandate “some sort of noise reduction” in phones, and most handsets have at least two microphones to enable that, he said. “Then it becomes a question of how low a power number can you achieve” while balancing performance and battery life, he said. The power profile required for noise cancellation “is a fairly well understood tradeoff,” he said.

But extra features -- such as noise reduction to improve accuracy with a speech recognizer when talking to a device to wake it up -- aren’t mandated by device makers, Alberty said. Enabling a phone to turn on by voice without a touch presents “a very deliberate tradeoff” in battery life because the phone has to be “always on” listening for “that special phrase or trigger” that makes it wake up and allow a user to interact by voice, Alberty said. “When our customers look at that kind of feature,” power usage becomes “really, really critical,” he said.

Cirrus addressed power consumption by optimizing its hardware and software with Sensory’s algorithm to “intelligently wake up the system in stages,” Alberty said. Once the chip detects noise it discerns the type of noise and whether it needs to listen for command phrases. The staged wake up system minimizes power usage throughout the process, he said, resulting in a power profile that results in “zero meaningful impact on battery life.” The solution also detects the voice of the person waking it up, so someone else can’t wake up a user’s phone even with the correct trigger phrase, he said.

This is the first Cirrus Logic “purpose-built” product that supports Sensory technology, Alberty said. The solution also works with Cirrus software that improves the accuracy of voice recognizers, he said. The noise reduction and signal pre-processing of ASR Enhance “make the signal as high-quality as possible” before it goes into a speech recognition engine leading to more accurate results when asking a voice recognizer like Siri for directions, he said. That can be the difference between getting directions to a Mexican restaurant two miles away versus “telling me to drive to Mexico City,” he said.