University of Washington researchers developed AI algorithms that enable people to choose which sounds they will hear through their headphones in real-time. They dubbed the artificial intelligence system “semantic learning,” which eliminates background noise by streaming recorded audio to a smartphone.
The most obvious application of this technology is enhancing noise-canceling audio equipment. However, it could help people spot specific things in their environment easier than ever. For example, an ornithologist or bird scientist could use these AI-powered headphones to ensure they can hear an endangered avian’s call and save it.
This article will explain how UW researchers developed their amazing AI headphones. Later, I will discuss other innovations in artificial intelligence and audio.
How did the researchers create AI headphones?
The UW scientists took various sound samples to create and train their algorithms. As a result, they allow a person to block 20 types of sounds, including:
- Bird chirps
- Sirens
- Human speech
- Vacuum cleaners
- Infant crises
Explaining this feature is simple, but developing it is far more complicated. Shyam Gollakota, the AI earphone study’s senior author, explained the complexity of creating their special sound accessory:
“Understanding what a bird sounds like and extracting it from all other sounds in an environment requires real-time intelligence that today’s noise-canceling headphones haven’t achieved.”
“The challenge is that the sounds headphone wearers hear need to sync with their visual senses. You can’t be hearing someone’s voice two seconds after they talk to you. This means the neural algorithms must process sounds in under a hundredth of a second.”
That time limit caused Gollakota and his team to design their semantic hearing system uniquely. It processes sounds on a mobile device like a smartphone instead of more powerful cloud servers.
Sounds from different directions reach our ears at different times. Consequently, the AI headphones maintain these delays to let users perceive sounds coherently.
The University of Washington team tested it in multiple offices like parks, offices, and streets. The device isolated sirens, bird chirps, and other target sounds in real-time.
They also had 22 volunteers rate the system’s audio output for its target sound. As a result, most said the quality is a significant improvement over the original.
However, Gollakota and his colleagues admitted their device had flaws. For example, the AI headphones struggled to recognize closely similar sounds like human speech and song lyrics.
Nevertheless, the researchers plan to resolve these issues by training their algorithms on more data. As a result, the semantic learning system can produce better results.
What are the latest AI audio innovations?
Many have been trying to blend artificial intelligence and music by pushing the limits of possibility. For example, Google is developing an artificial intelligence called MusicLM that creates songs from text.
It uses machine learning and algorithms to create new content from a database. The latter only has 280,000 hours of music, which is relatively smaller than Spotify’s humongous library.
Nonetheless, it combines music with surprising variety and depth. For example, it could create a hybrid of K-pop and Classical with a “spacey, otherworldly” tune that promotes a “sense of wonder and awe.”
MusicLM can also create melodies based on whistling, humming, or the description of a painting. Moreover, its story mode can merge several descriptions to create a DJ set or album.
However, MusicLM has numerous flaws, like ChatGPT and DALL-E. The most apparent one is that some of its compositions may sound weird or incomprehensible. It can also sound more repetitive than manmade tunes.
Another research team from the University of Lincoln has been trying to create an AI system that lets us understand cats. They started by creating a database with millions of expressive cat photos labeled with corresponding emotions.
Their artificial intelligence uses those samples to estimate what felines feel. “It could highlight the rules it uses to distinguish data sets, which can show us where to look for the best way to distinguish certain expressions,” said Daniel Mills, a veterinary behavioral medicine professor at the University of Lincoln.
Conclusion
University of Washington scientists created headphones that let people hear what they want to hear. For example, users may mute all background noise except for birds chirping.
As a result, future earphones could deliver sound quality far superior to previous ones. Moreover, AI headphones could become essential for research and other purposes.
Learn more about this experimental audio equipment in the Association for Computing Machinery’s Digital Library. Also, check out more digital tips and trends at Inquirer.