IoT developers face rising demands for touchless technologies like voice control.
In a post-COVID world, Internet of Things (IoT) developers face rising demand for touchless technologies—like voice control.
Vikram Shrivastava, senior director of IoT Marketing at Knowles, noted that “with COVID, we’ve seen an uptick on Bluetooth devices. The main reason is that a lot of people working from home are using these speakers as conference speakers—they pair it with their phone and laptop.”
It makes sense. Adding voice control to IoT devices, both at the consumer and the enterprise level, gives the impression that technology is protecting us from spreading illness via contact. This impression increases demand.
The challenge in meeting this demand is that the IoT market is fragmented. How you integrate voice control into each application—be it Bluetooth speakers, home appliances, factory equipment or elevators—will be different. Shrivastava explains that adding a voice wake trigger can be simple, but that designing an enterprise-grade Bluetooth speaker and headset is a lot more complex. If that speaker includes true wireless stereo (TWS) integration, the complexity rises once again as you enter a different power envelope.
Additionally, various applications necessitate voice integrations with different ecosystems. For instance, you need to work in a Linux ecosystem to implement voice on most smart TVs, but to get voice on a home appliance will require working in a microcontroller (MCU) ecosystem. Enter Knowles’ standard voice integration initiative:
“The goal of the Knowles industry standard solution is to have preconfigured starting points for a customer, system integrator or OEM,” said Shrivastava “They can then take that solution and start prototyping [for any scenario].”
Why Knowles Believes It Can Create a Standard Voice Control System for IoT
Knowles has been in the audio space for over 70 years. Back in the 1940s it started in the hearing aid business. Since then, it has made history by having technology on lunar landers and Mars modules.
Currently, the company is known for its micro-electro-mechanical systems (MEMS) microphones, which can be found in billions of mobile phones, laptops, IoT devices, headsets and in-ear devices.
But Knowles isn’t just a component provider. Five years ago, it acquired digital signal processing (DSP) technology that enables it to provide software, algorithms and signal processing to handle the data collected by its microphones. Knowles also has labs that study acoustics and tune product performance.
Shrivastava explains that this approach has positioned Knowles to become a solutions provider—specifically one aiming to develop a standard voice control system for connected devices. He said, “We’re enough into the solution space where we can build Linux or android-based software to integrate our DSPs and get voice running on different platforms as fast as possible.”
What Would a Standard IoT Voice Control System Look Like?
“What do we call a standard solution? We look at what would be the common denominator and requirements for a vertical, so that we release not just an SDK but a system level solution, or components that enable a system level solution,” said Shrivastava.
Recently, Knowles released a voice control solution for IoT devices that include a Bluetooth system on chip (SOC) as a host. The solution doesn’t have a heavy ARM core, it doesn’t run Linux, and it doesn’t run Android. But it is a standard solution for the Bluetooth and Bluetooth Low Energy (BLE) verticals. This opens the door to adding voice control to a vast number of IoT and smart home devices.
As Shrivastava said, the goal is to provide standard preconfigured starting points, not a universal solution for every IoT device. With respect to anything that runs Bluetooth, this broad market development kit can be a standard solution. As Knowles aims to produce more standards, expect to see similar development kits that can add voice controls to other IoT ecosystems.
Shrivastava explained, “[to] offer something across the board that can be used for something that has a Bluetooth engine, or something that has a very simple MCU, or an application processor that’s running android—in each case you have to put a separate set of system solutions that help the developer of the device to bring voice into that ecosystem as quickly as possible.”
Why Should Standard Voice Control Solutions Interest Engineers?
The benefits of Knowles’ standard solution are that it speeds up the time to market for any Bluetooth device that implements voices controls. It also expands the technical capabilities and sophistication of these devices in a way that can be affordably scaled to an individual product’s demand.
Shrivastava noted that on average, “Bluetooth chips are fairly limited. They have very simple audio capabilities. But what if you are designing a speaker or a headset that needed Microsoft Teams certification? That is a significant level of performance that Bluetooth chips are not capable of. That’s at the heart of our reference solution. It includes Knowles low power DSP, which enables these high-end features—which would be difficult for a Bluetooth chip to achieve on its own.”
Engineers will also be able to use the standard solution to affordably implement other functionality. The solutions, audio front end and algorithms handle the physics and acoustics behind integrating voice control into an IoT device. This front end is capable of:
- Acoustic echo cancelation: to eliminate speaker noises that the microphone picks up
- Beamforming: to improve voice capture
- Noise suppression: to clean the captured signal
Once the recording is clean, it is sent to the voice trigger. Knowles’ works with various partners, like Amazon and Google, for trigger word development. As a result, much of the hard work behind the hardware and software of voice control is done for the engineer.
The Machine Learning Possibilities of Voice Control
Knowles DSP is also compatible with machine learning capabilities for audio event detections. Shrivastava said, “We could listen to a sound and say, ‘that’s glass breaking.’ We run a sound classifier engine that is extremely efficient; it’s in the one milliwatt range.”
Theoretically, this machine learning capability could listen to its surroundings, sense something is wrong, and warn people of potential dangers, complications or issues.
For instance, a sensor in a biohazard lab can listen for the sounds of a vial breaking. It could then ask the technician to confirm if they are safe. If the technician answers yes, then nothing happens. If the technician calls for help, or makes no reply, the system could warn others of a potential hazard.
Similarly, the sensor can be trained to listen to the noises of factory legacy equipment. If it senses odd noises from that 30-year-old industrial mixer, the sensor could be trained to send warnings to maintenance workers.
The examples of a sensor’s uses don’t have to be so dramatic. Shrivastava imagines a sensor on a dryer that listens to the noises produced by the drum. If that sensor hears that something is off, it could, for instance, send a message to a user’s cellphone to lighten the load in the dryer.
In short, with a simple Bluetooth communication interface, engineers can utilize Knowles’ DSP to design products that understand commands, on the network edge, without sending data to the cloud. This can significantly reduce the bill of materials and power supply traditionally needed to perform these functions.
