Adding low-power voice recognition technology to TV remotes

“Virtual assistants are being integrated into more devices in the home, with Amazon recently announcing that it will introduce a variety of new devices for Alexa, including earbuds, glasses and rings, to give consumers more access to information. A recurring theme in these new voice products is being wireless and hands-free, making it easy to connect to a cell phone or other host computer and listen to commands. The technology inside is a Bluetooth RF chipset that supports wireless connectivity and a dedicated embedded processor running a Wake Word Engine (WWE) to recognize voice commands.

“

Virtual assistants are being integrated into more devices in the home, with Amazon recently announcing that it will introduce a variety of new devices for Alexa, including earbuds, glasses and rings, to give consumers more access to information. A recurring theme in these new voice products is being wireless and hands-free, making it easy to connect to a cell phone or other host computer and listen to commands. The technology inside is a Bluetooth RF chipset that supports wireless connectivity and a dedicated embedded processor running a Wake Word Engine (WWE) to recognize voice commands.

Another example of this trend is that every new flat-screen TV, set-top box and media player comes with a remote control, and they will soon be completely wireless and hands-free too. Of course, a lot of people still use the old infrared sight remote mode or key-to-talk when you want voice control, but those are quickly obsolete. Users want a device that responds seamlessly to commands, rather than having to press and hold the talk button like a walkie-talkie. Similar to the recent wave of Amazon’s electronics, the next generation of TV remotes will be wireless and hands-free.

However, remote control design also presents some unique challenges. For example, remotes are usually not rechargeable; they usually use standard AA batteries. Not only does the remote need to perform well in a noisy environment, but it needs to be 3 to 9 feet away from your body, and wirelessly and instantly transmit information to a host device (such as a TV).

Additionally, consumers prefer durable batteries that don’t need to be swapped out frequently, and the remote must essentially perform like the Amazon Echo Dot while consuming less power than in-ear headphones.

Designing a powerful, energy-efficient remote control is a challenge that requires innovative Bluetooth solutions and audio processing solutions, as each of these improvements extends battery life.

Using Bluetooth 5.0/LE solves several problems of traditional infrared. First, Bluetooth is a standards-compliant solution, so devices can easily communicate with existing Bluetooth devices. Additionally, Bluetooth 5.0/LE offers a range comparable to WiFi devices, which is great for voice remotes. Traditional Bluetooth solutions have been optimized for phones and laptops, which have larger batteries and don’t consume as much power as Bluetooth.

Atmosic’s innovative solution can significantly reduce power consumption by 5 times and extend battery life by 3 to 5 times compared to competing solutions.

In addition to the very low-power Bluetooth design, it is also possible to use a secondary wake-up receiver that can use less power (20 to 50 times less than a standard receiver) to put the entire Bluetooth SoC into a deep sleep state; the The device can be woken up by a special mode of another host. We won’t go into the details of this technology here, as it only works with a small number of dedicated remotes.

A third low-power technology utilizes energy harvesting to harvest RF wireless energy to extend battery life. Many homes and buildings have large amounts of RF energy (usually in the ISM band) that can be harvested when the remote is placed on a table. Depending on the energy level, a device can obtain tens of microwatts of energy. The ultimate goal is to replace battery power where possible, extending the life of the battery to several years, compared to the 6 to 9 months of the current remote control. For industrial and special-purpose remote controls, other energy technologies such as solar, thermal and motion energy harvesting are also available.

As mentioned earlier, for true hands-free operation, the remote must simultaneously work like a smart speaker, but also be as energy-efficient as in-ear headphones. Companies like QuickLogic have developed highly optimized, ultra-low-power companion devices to accompany Bluetooth chipsets to address this challenge.

The Bluetooth-connected voice remote control basically has three modes: standby mode, wake-up word detection mode and data transmission mode, and the power consumption is also gradually increased.

In the standby state, the Bluetooth and companion chips are dormant, waiting for some wake-up words from the surrounding environment. One of the most energy-efficient ways to achieve this is to use Vesper’s Microphone Wake-up Sound feature, which consumes only 10µA and waits for ambient sound to exceed a pre-configured threshold. In a typical living room use case, the system is in this mode 80% of the time.

Once the threshold level is reached, an interrupt is triggered from the microphone and wakes up the companion chip; at this point wake word detection mode is entered. The MCU that accompanies the chip can boot and run WWE for a period of time to detect if a keyword is spoken. Third-party solutions such as VoiceSpot WWE from Retune DSP can run on Cortex-M4 using only one microphone, eliminating the need for computationally intensive solutions using multi-microphone adaptive beamforming, which is typically Required for voice recognition in midfield (3 to 9 feet).

In addition to the obvious savings in processor resources, removing each microphone from the system saves 400 to 650 µA (active power). If a wake word is detected, it interrupts and wakes up the bluetooth chip into data transfer mode. This is necessary because the user word after the wake-up word needs to be delivered to the host (eg TV) in the form of pulse code modulation (PCM) or compressed data.

If no wake word is detected, the system will revert to the initial standby mode. Some companion chips, such as those from QuickLogic, have dedicated low-power sound detection (LPSD) hardware to reduce the average system power used in wake word detection mode. For example, some fan sounds have a high decibel sound pressure, but are clearly not speech. The LPSD hardware is smart enough to sense this and ignore the sound to avoid running WWE unnecessarily for extra power.

Bluetooth 5.0/LE is ideal for data transfer mode because it can transfer data in low-power on-demand packets. An ideal companion should have enough storage and processing power to compress voice data before sending it to a Bluetooth device, a typical example is running the Opus audio encoder.

Author of this article:

Scott Haylock: Director of Product Marketing at QuickLogic. He has over 20 years of system-on-chip experience and holds a bachelor’s degree from Michigan State University.

Srinivas Pattamatta: Vice President of Business Development at Atmosic Technologies. He also has over 20 years of experience in wireless and other communication technologies. Received a master’s degree in electrical engineering from Oregon State University and a master’s degree in business administration from Santa Clara University.