Let’s Keep It Between Us

With the help of modern technology, muting distracting voices in video calls has become a simple process. With just a few clicks, you can silence unwanted background noise and side conversations, ensuring that the primary speaker's message remains clear and uninterrupted. This is largely possible due to the sophisticated audio processing algorithms and noise cancellation features built into video conferencing software. These technologies work together to distinguish between your voice and the surrounding sounds, allowing you to have focused and productive virtual meetings.

However, when it comes to recording real-world interactions where multiple people are present in the same physical location, replicating the same level of control over distracting voices becomes an entirely different challenge. Unlike the controlled environment of a video call where each participant has their own microphone and audio feed, physical gatherings often involve shared spaces and ambient noise. Traditional recording equipment, such as handheld microphones or room microphones, captures all sounds within their range indiscriminately. This means that in a real-world meeting, it is nearly impossible to selectively mute or isolate specific voices.

One can take precautions to minimize background noise and encourage attendees to be considerate, but there is no easy technological solution to separate voices cleanly from a shared physical space. That may change in the near future, however, thanks to a swarm of robotic smart microphones that was recently developed by researchers at the University of Washington. The robots autonomously navigate around a room to distribute themselves optimally, then they run onboard deep learning algorithms to identify and track the position of each individual speaker.

The robots are powered by Nordic Semiconductor nRF52840 microcontroller, with a Bluetooth Low Energy module added for wireless communication capabilities. By selecting a sufficiently powerful microcontroller, the algorithms can run on-device, sidestepping any privacy concerns associated with sending a stream of audio to a cloud service. A gyroscope and accelerometer provide odometry information, and a pair of micro motors provide for locomotion. A pair of microphones and a speaker are onboard, and the entire system is powered by a rechargeable LiPo battery.

Each of the circular, wheeled robots is about an inch across. After deployment from a charging station, they all travel in different directions, emitting high frequency sounds to communicate with one another and make sure they are spread out as far as possible from one another.These high-pitched chirps also keep the robots from falling off of tables or bumping into other obstacles.

By feeding audio information into a deep neural network, including the delay in time with which a particular voice arrives at microphones in different positions, the team found that it was possible to locate and track individual speakers in space. And the algorithm proved to be accurate enough to distinguish between speakers, even if they had similar sounding voices and were near to one another. Experiments showed that the present system was able to accurately distinguish between nearby individuals 90% of the time. Importantly, this was accomplished without prior information about the individuals, so no initial training session is required to calibrate the device.

The capabilities of the robot swarm could be leveraged immediately to mute the microphones nearest to a distracting voice, even if that individual is moving around. As a next step, the researchers would like to explore using their techniques to create “active” zones around voice-controlled devices where commands can be accepted, with the hope that more distant interfering sounds could be ignored. And further down the road yet, they plan to explore incorporating noise-canceling technologies into the robots. That addition could allow them to mute noisy areas in physical locations, producing a sci-fi-like cone of silence.