Bixby Wake‑Up and Unlock System (Speaker Recognition)

Speaker‑recognition–based wake‑up and unlock system for Galaxy devices, enabling hands‑free activation of Bixby and secure voice‑based unlocking.

Overview

Hands-Free Convenience

Galaxy device users are used to asking Bixby to help them search the web, make a reservation, or launch an app when their hands are full. Now, Bixby’s new Voice Unlock feature allows you to activate and interact with Bixby when your phone is locked for an even more comfortable user experience.

Bixby Voice Unlock creates a convenient way to utilize Bixby when your phone is out of reach. Say you’re in the middle of making dinner and you need to double check exactly how much of a particular ingredient the recipe calls for. With Bixby Voice Unlock, rather than smudging up your screen with sauce and bits of food, you can simply say “Hi Bixby” to unlock your phone, and then ask Bixby to pull up the recipe for you.

Contributions

July 2018 - August 2019

I designed and implemented the core software architecture for the Bixby wake‑up and unlock system used in Samsung Galaxy devices. This system uses speaker recognition to recognise a user’s voice and activate Bixby hands‑free, allowing the phone to wake up and unlock when the enrolled user utters a wake‑up phrase.

I developed robust algorithms to improve recognition accuracy across various noise environments and implemented mechanisms to update user enrolment data for long‑term reliability.

  • Core architecture: Built the speaker‑recognition pipeline, handling audio capture, feature extraction, speaker embedding and decision logic.

  • Noise‑robust recognition: Developed algorithms that maintain high recognition accuracy across noisy environments by employing adaptive filtering and robust embedding models.

  • Enrolment data update: Implemented methods to update and refine enrolled speaker embeddings over time, allowing the system to adapt to changes in the user’s voice and maintain security.

  • Registration for speaker recognition (US11545154B2) – Describes techniques for registering a user’s voiceprint and updating enrolment data to improve recognition accuracy over time.

  • Speaker recognition registration (US11289098B2) – Provides methods for capturing, storing and managing speaker templates used in wake‑up and authentication.

  • Data recognition (US11574641B2) – Covers a framework for recognising patterns in sequential sensor data, including speech, which supports the robust handling of voice signals in the Bixby system.

  • Selective combined authentication (US11625473B2) – Introduces a mechanism to combine different biometric modalities, enabling voice‑based authentication to be selectively augmented with other modalities when needed.

These patents capture key techniques behind Bixby’s wake‑up and unlock system, including voice enrolment, continuous adaptation and multi‑modal authentication strategies.

[1] https://news.samsung.com/global/in-depth-look-4-more-powerful-more-personal-bixby-on-the-galaxy-s10