Researchers addressed the longstanding “cocktail party problem” by building a prototype called proactive hearing assistants. The device uses AI to follow the rhythm of a conversation and to highlight the voices of the people you are speaking with. The team presented the work in Suzhou at the Conference on Empirical Methods in Natural Language Processing and made the code open‑source.
The system has two models. The first determines who spoke when and looks for low overlap in turns. The second isolates the identified speakers and reduces other voices and background noise. The models can identify partners with only two to four seconds of audio and can handle one to four partners besides the wearer.
The team tested the headphones with 11 participants. Users rated the filtered audio more than twice as favorably as the baseline. The researchers said challenges remain with overlapping talk, long monologues, and people entering or leaving conversations.
Difficult words
- prototype — early model of a new device
- proactive — acting before a problem happens
- rhythm — regular pattern or beat in speech
- isolate — separate one sound from other soundsisolates
- overlap — when two people speak at same time
- baseline — standard condition used for comparison
- monologue — a long speech by one personmonologues
Tip: hover, focus or tap highlighted words in the article to see quick definitions while you read or listen.
Related articles
Algorithms show how catalysts turn propane into propylene
Researchers at the University of Rochester developed algorithms that explain how nanoscale catalysts convert propane to propylene. The work reveals atomic features of metallic and oxide phases and could help improve industrial production methods.
Reliable Internet Unavailable for 90% of Poorest
A report highlights that 90% of people in low- and middle-income countries lack reliable internet access, affecting their ability to access essential services.
Missing real‑world detail drives AI bias
Research from the University of Texas at Austin finds that AI bias often comes from models that miss complex real‑world conditions. The study of 363 algorithms points to three key risk factors for unfair outcomes.
Small pause to slow misinformation on social media
Researchers at the University of Copenhagen propose a small pause before sharing on platforms like X, Bluesky and Mastodon. A computer model shows that a short delay plus a brief learning step can reduce reshares and improve shared content quality.
Coho salmon DNA detected in air at Issaquah Creek
Last fall researchers collected Coho salmon DNA from the air beside Issaquah Creek, near a salmon hatchery in Washington. They combined air and water eDNA with hatchery counts to track the salmon migration.
People with AMD Judge Car Arrival Times Like Others
A virtual reality study compared adults with age-related macular degeneration (AMD) and adults with normal vision. Both groups judged vehicle arrival times similarly; vision and sound were used together, and a multimodal benefit did not appear.