One of the most unique audio applications introduced at CES this year was Vogtec’s Cybic AI bicycle, equipped with a built-in voice user interface (VUI). Using the VUI and local and cloud-based technology, a rider can make hands-free calls, communicate with other riders, and interact with their bike to plan routes and track progress.
At the core of the VUI is the SmartVoice SoC with HD Clear voice processing from DSP Group, which is aiming at the market for audio applications with a suite of hardware, software, and data compression solutions.
“A bicycle outdoors is a very challenging environment for a voice application,” said Avi Keren CTO and Head of DSPG Software R&D and SmartVoice Division Manager, DSP Group. “But here we were able to overcome ambient noise and ensure that the system only receives the rider’s voice, through a combination of our proprietary far-field ,multi-microphone algorithm along with beamforming, noise reduction, and acoustic echo cancellation.”
From bikes to hearbuds
With a focus on the optimization of its technology for low power, DSP Group engineers discovered that AI can be used very efficiently for a variety of voice command and amplification applications. All of the AI processing is done locally in our chip.
For example, the company’s SmartVoice technology—an advanced low-power voice processing platform and algorithms-- is also at the center of Google’s Pixel Buds 2. In this growing category of consumer electronics, Technics and Panasonic similarly have announced new headsets with an active noise-cancellation (ANC) capability from DSP Group.
“Most of the ANCs implement by hardware, but we have a very advanced algorithm and are basically using AI to characterize the sound field around the wearer and do the ANC autonomously,” said Keren.
Smart homes getting smarter
Home security technology that relies on sound event detection is next on the horizon. Ideal for this application, said Keren, is a new multicore SoC that DSP Group will shortly be announcing. It’s a dedicated neural network processing engine focused on doing deep neural networking at very low power consumption.
“Using AI, this technology will be capable of detecting with a high degree of reliability sounds such as breaking glass that would indicate an intrusion or other scenario for which you might need to request intervention by a security company,” said Keren. He noted that such a device would feature a VUI and the capability to support a phone call to an emergency center if required.
In order to build up a library of recordings and sound vectors to train the AI system, the company is currently capturing as many sound vectors as possible. “We have acquired sounds of glass breaking, alarms for everything from carbon monoxide to fire detection sensors, and clips of these sounds in different environments, because, for example, an alarm going off in a big room compared to a small room will be different,” said Keren.
For engineers dreaming up the next audio application leveraging AI technology, Keren recommends that they pay close attention to the acoustic design, including the placement and isolation of microphones. “Where customers have problems, it is usually because they didn’t do the acoustic design right from the onset,” he noted.
When choosing an SoC, he said that while meeting the requirements of the application is a given, engineers should look for as integrated a solution as possible in order to save on costs. Opting for an SoC with both analog and digital microphone interfaces reduces the number of components.
DSP Group's Avi Keren is participating in a panel discussion on AI in Consumer Electronics, on Monday, August 10, 2020 during Fierce AI Week, a free virtual event. Visit the event website for more details and to register for this event running August 10-12, 2020.