Knowles Corp. has announced the first development kit in its Intelligent Standard Solutions portfolio, which is designed to help developers get to the prototyping stage very quickly and accelerate their time to market. The new AISonic Bluetooth Standard Solution is a complete development kit that enables voice integration into any Bluetooth-connected device.
With this solution, OEMs and ODMs can build voice-activated calling, voice control, and far-field speech recognition capabilities into their Bluetooth device designs. These include smart speakers, smart home locks, connected light switches, wearables, and in-vehicle voice assistants.
The new portfolio of reference solutions – known as the Knowles Intelligent Standard Solutions – are comprised of DSP processors, software, and integration support for voice activation, voice control, and contextual audio processing for a range of devices. The AISonic Bluetooth dev kit is available now. Voice solutions for TVs, sound bars, and white goods are expected to follow later this year.
One key reason why Knowles launched the standard portfolio starting with Bluetooth speaker voice integration is increased demand. “With Covid-19 there are a lot of people working from home and they are beginning to use Bluetooth speakers as their conference speakers,” said Vikram Shirastava, senior director of IoT at Knowles. “This has increased demand for voice calling capability on these Bluetooth speakers.”
But beyond Bluetooth speakers, the demand for voice is showing up in all kinds of applications and verticals, such as using voice for elevator control and smart home devices, and integrating voice into all of these devices is not very easy, he said.
Shirastava said there are a lot of things to consider in these designs such as microphone placement, acoustics for the use case, and specific requirements of the device itself. For example, a tablet uses a very powerful applications processor while white goods like a microwave use entry-level microcontrollers (MCUs) that just control buttons, so there is a whole range of challenges, depending on the application, making the market very fragmented, he added.
With the standard solutions, Knowles is trying to find the common denominator to bring voice to specific verticals with the SDK and software components, said Shirastava.
“Apart from being a components supplier, we are becoming a full solutions provider and that includes the software that’s required to integrate our DSPs with an MCU or with an application processor in a given use case, understanding the algorithm that is needed for that voice use case, and also being able to tune those algorithms to work with the acoustics of the device,” said Shirastava. (Knowles, a leading manufacturer of MEMS microphones, added audio DSP technology to its portfolio via its acquisition of Audience Inc., a developer of audio algorithms on DSPs, in 2015.)
He noted that Knowles is addressing scalability through a standardized approach. The standard solution approach is not exactly turnkey, which is “scalability through inflexibility because you’re not allowed to change components,” he said. “You have to take it as is and that is how turnkey works, but voice cannot be made turnkey.”
Instead Knowles’ scalable system level solution takes a framework approach. Each solution will take multiple factors into consideration based on the end application. In the case of Bluetooth speakers, as an example, this means the different Bluetooth chipsets, operating systems, microphone performance, and algorithms.
There are also a variety of other factors. “Are we going to be working with voice UI, is it with calling enhancement, is there a cloud ASR [automated speech recognition] or is everything going to be at the edge, and that puts together a framework on how to go in and engage with Bluetooth speakers,” said Shirastava.
The Bluetooth kit goes beyond just being a simple headset application, there is a lot more built on top of this at the system level, said Shirastava. These include understanding the API, standardizing the acoustics performance, and at the same time standardizing with two or three or four mics, and doing the audio processing chain, which includes acoustic echo cancellation (AEC), noise suppression, and cleaning the audio signal, he said.
What’s in the dev kit
The AISonic Bluetooth Standard Solution includes the reference board with the Knowles IA8201 audio processor, SDK, pre-integrated microphones, an API that supports any Bluetooth chipset with an embedded MCU running an RTOS, and system firmware configured to support sensors. At the center of the kit is the AISonic audio edge processor (IA8201), claiming the industry’s lowest power consumption to enable wake-on-voice functionality in small battery-operated devices.
The IA8201 combines two Tensilica-based, audio-centric DSP cores. One core targets high power compute and machine learning (ML), while the other one is used for very low power always-on processing of sensor inputs.
Knowles IA8201 board (Image: Knowles)
“Our DSP as a front-end has the capability of meeting the Amazon AVS [Alexa Voice Service] requirements at the edge,” he said. “When you look at Arm cores and SoCs in a typical [Amazon] Echo device, they are multi-cores running at 1-GHz and they consume almost one watt running the audio front-end engine. We added almost 400 new instructions to our DSP, also built on a multi-core, that specifically accelerates audio and and machine learning, so it takes us only about 40 MHz.”
That is how the DSP achieves a significant power consumption savings and why now a Bluetooth device, which is battery-powered, can be always listening, he added. “In the past, a lot of the Bluetooth devices had push-to-talk capability where you actually had to push a button on the Bluetooth device in order for the microphone to listen to you but since we can do always listening at less than 15 milliwatts, you don’t have to have a button anymore, it can be always listening for the hot word and and doing all of that audio front-end processing.”
In terms of ML, a lot of the trigger word algorithms have moved towards using some neural network capability, but Shirastava said Knowles uses it for a lot more than that, such as noise source classification. “We classify the noise source using machine learning algorithms and then tell the noise suppression algorithm exactly what to do. This reduces a lot of the artifacts that noise suppression algorithms do to our voice before it goes to the trigger engine.”
The AISonic Bluetooth Standard Solution includes algorithms for an audio front-end suitable for use with many voice assistants or cloud-based ASR APIs. In addition, Knowles has an OpenDSP partner program that allows many algorithm developers to provide libraries for beamformer omnidirectional voice capture, and AEC to sort voice commands from the background to meet the false wake standard of AVS.
One partner is Retune DSP, which has ported its voice-control solutions, VoiceSpot and VoiceSeeker, to the IA8201. The VoiceSpot, an ultra-compact wake word engine, enables voice-wake on memory-constrained, always-on, battery-operated edge devices, and is Amazon-certified. VoiceSeeker is a multi-microphone audio front-end preprocessing solution capable for far-field voice control applications. In addition, Retune’s Conversa two-way voice communication software solution enables the IA8201 on a number of high-performance Bluetooth speakerphones and headphones.
Reducing development time
Knowles cited a few examples where the voice integration solution has helped its partners develop AWS-AVS, Alexa Auto, and AVS-AMA solutions on several Bluetooth and embedded MCU platforms.
One example is Sugr, a solution provider of voice algorithms and an integrator, which recently completed the iOttie Aivo Connect integration with the Knowles IA8201. Sugr’s DSP algorithms and system-level capabilities are also part of Knowles’s AlSonic Bluetooth Standard Solution, helping OEMs get certified with AVS.
iOttie Aivo Connect (Image: Knowles)
The first product to use the standard solution is the iOttie design that is not just an AVS-AMA device but it is also certified for Alexa Auto, said Shirastava. “It’s basically a wireless car charger with two microphones in the base so when you power up your phone and launch the Alexa app it will find this car charger as an Alexa device, allowing you to add it as an Alexa Auto device. So now in your car you can have Alexa paired all the time.”
“We develop the audio front-end – voice algorithms – to help products to clearly record the human voice and to reduce environmental noise, echo noise, and other reverberation noise, etc., to make sure the human voice recording is clear for either human-to-machine communication or human-to-human communications,” said Sean Song, Sugr’s CEO,
Sugr also integrates its technology into consumer electronics products for its customers such as iOttie, a design-centric product company in North America.
The Knowles standard solution streamlined the traditional DSP development, integration, and optimization, said Song, with the framework enabling its engineers to develop the SugrSense Bluetooth AMA module with the IA8201 efficiently.
“Knowles has a very good product, which is the the DSP chipset, and they also provide to their partners like us a very good framework, which makes our development easier and faster which means more proficient,” he added.
One of the biggest benefits of Knowles’ standard solution, said Song, is the savings in development time. “Knowles offers a framework, which provides optimization libraries and connectivity libraries, so our engineers can build our blocks based on that.”
“As one example, we are an algorithms-centric company, which means if we finish one second of voice processing it takes lots and lots of computation and if we do that by ourselves it takes about a month to optimize the code, but Knowles provides a standard library that we can almost instantly use to reduce our engineers working hours,” said Song.
Knowles also provided the Bluetooth connectivity reference design, which further reduced our development work, he added.
Advertisement
Learn more about Knowles
