[Maui, United States=Digital Today reporter Seok Dae-geon] On-device AI in smartphones has advanced to the point where it can remove shouting and clapping from a few steps away and send only the voice of the person speaking in front to the call recipient. In a demo room at Snapdragon Summit 2026, held in Maui, Hawaii, from Sept. 22 to 24 local time, sound AI sent Korean and English spoken within a 1-metre radius in front of the phone to the other side's speaker, while sounds made outside that area were not heard. The feature that filters and listens to a voice was ready for immediate use in real calls.

A demonstrator introduced the feature as an "AI voice bubble". The smartphone's microphone and on-device AI create a space of about 1 metre around the device and accept only the voice spoken inside it, the demonstrator said. The setup placed a call from one smartphone to another, with the receiving side connected so the sound came out through a Bluetooth speaker behind the booth.

As the demonstrator kept speaking and stepped back, the voice coming from the speaker cut out. The speaker stayed quiet even when someone shouted louder and clapped outside the bubble. Sound enters the microphone, but the AI ignores it. When the speaker moved back in front of the smartphone, the voice was delivered again immediately. Noise made by a hand while speaking inside the bubble was also filtered out, and only the voice reached the other side.

When "Hello, nice to meet you" and an English test sentence were spoken alternately in front of the smartphone, the voice came through the speaker as it was. The demonstrator said, "It is capturing and delivering only your voice." The voice inside the bubble did not cut out while Korean and English were mixed over three rounds. For the listener on the other side, only the speaker's voice remained instead of the booth's chatter.

◆Calls, dictation, translation and conversations with AI assistants all affected by voice quality

The smartphone determines how far away a speaker is using differences in the time it takes for sound to reach multiple microphones. It calculates the time it takes sound to reach each microphone to create an area about 1 metre across and responds only to voices within it. In the demo, even a louder voice shouted from outside the bubble did not pass through. On-device AI is responsible for judging and distinguishing voices from other sounds.

Qualcomm plans to connect the voice-filtering technology to agentic AI. As voice becomes a main way to call AI assistants and agents, accurately capturing a voice has become more important, Qualcomm said. That means calls, message dictation, conversation translation and conversations with AI assistants are all affected by sound capture quality.

Qualcomm also added AI voice technology to Qualcomm CVC in its audio platform, Snapdragon Sound Elite 2nd Generation. It also presented a case where an agent connects to related services and completes tasks when a user, while wearing earphones, asks by voice to find a cheaper price or change an appointment schedule. It also added a function that judges surrounding sound. If a siren approaches while walking in a city with noise cancelling on, adaptive ANC recognises it and switches to transparency mode to let the sound through.

The name AI voice bubble is not an official title yet. Asked why it is called a "bubble", the demonstrator said it is an internal name used to explain the feature easily, adding, "It will have another name when it comes to market." The demonstrator added, "In the end it depends on how OEMs implement it, and each will attach the name it wants when putting the technology into a device."