Researchers at the University of Washington have created AI-powered headphones that can automatically recognize speakers and mute voices that don't match the rhythm of the conversation. In the future, the technology could be scaled down to the size of in-ear headphones or hearing aids.
Researchers at the University of Washington have created AI-powered headphones that can automatically recognize speakers and mute voices that don't match the rhythm of the conversation. In the future, the technology could be scaled down to the size of in-ear headphones or hearing aids.
The technology is built on two AI models: one that analyzes time patterns and the other that filters out unrelated sounds. The system identifies speakers within two to four seconds, Interesting Engineering writes .
“Our understanding is that when we talk to a group of people, our speech naturally follows a rhythm of alternating lines. And we can train AI to predict and track these rhythms using only audio, without the need for implanted electrodes,” explains the team’s senior researcher Shyam Gollakota.
The system is activated as soon as the wearer of the headphones starts talking. The first model runs a “who spoke and when” check, looking for minimal voice overlap. The second model cleans the signal and returns isolated sound to the user in real time.
The models have been trained in English, Chinese, and Japanese. Other languages may require additional settings.
The prototype currently supports conversations between the user and up to four other people without noticeable lag. The researchers tested it with 11 participants.
The team also showed that similar artificial intelligence models can already run on chips the size of a hearing aid.
