New kind of AI audio chip transforms earbuds into always-on 'intelligent' devices that run without a smartphone. How does it work?

MAUI, Hawaii — Engineers have built a first-of-its-kind artificial intelligence (AI) audio chip for wearable devices that they say will let users engage with AI assistants without needing to maintain a smartphone connection. Devices with the new chip will also harvest audio ‪—‬ and even video ‪—‬ data from the world around you to better serve these virtual agents.

The new chip, called Snapdragon Sound Elite Gen 2, will pave the way for new features that are currently unavailable directly in portable devices such as headphones, smart glasses and speakers. Qualcomm representatives announced the new chip at the company's annual Snapdragon Summit, hosted here Sept. 21-24.

Unlike current earbuds, devices with the new chip will be able to host apps right on the memory of the devices themselves, without relying on a continuous smartphone connection, as well as sync with cloud-based services like ChatGPT when connected to Wi-Fi. In practice, the chip will enable users to make vocal queries and requests, with an AI model taking actions such as setting up entries in an online calendar or making online purchases. Unlike existing devices that can perform similar functions, this would be possible without a smartphone connection.

Beyond earbuds, company representatives want to integrate the chip into other "hearables," such as over-ear headphones, open-ear audio devices, audio glasses, camera buds (headphones that use tiny built-in cameras) and hearing devices, as well as "future personal AI wearables," like pendants or clips.

For users who wish to connect hearables to a smartphone or computer (to interact with a visual user interface, for example), the new chip uses a specialized component within the platform to establish connections over micropower Wi-Fi channels alongside Bluetooth. This technology harnesses architectural improvements at the chip level to consume much less power than standard Wi-Fi connections — but at the expense of higher latency and lower peak speeds. For audio devices, however, using Wi-Fi rather than Bluetooth generally improves playback quality, given that Bluetooth compresses data during transmission, Qualcomm representatives said.

Adaptive and selective active noise cancellation also means users can filter out audio on command. For example, it allows listeners to hear key sounds, such as emergency sirens, while walking through a city or highlight people's voices during a conversation, while blocking out everything else. This noise cancellation is achieved through standard machine learning algorithms, rather than transformer-based large language models that power AI systems like ChatGPT.

Devices could also allow lossless audio streaming — a high-quality audio standard that's usually available only if you subscribe to premium streaming services, like Spotify or Tidal — thanks to another built-in subsystem.

Qualcomm representatives also highlighted the chip's "always-on intelligence," in which an on-device eNPU AI accelerator component powers virtual assistants, transcription and translation, as well as the building of "contextual awareness." With contextual awareness, the system would constantly monitor the user's environment to harvest data to add to a personal profile, which representatives say helps it answer future queries.