Samsung Bixby Can Deepfake Your Voice to Answer Phone Calls
Samsung Bixby device owners will be able to teach the voice assistant to deepfake their voice when answering phone calls. The Custom Voice Creator feature of the Text Call service arrived as part of a package of upgrades to the voice assistant and its AI abilities.
Samsung introduced the Bixby Text Call feature last year as a text-to-speech and speech-to-text tool for answering the phone. Users activating the feature see what the caller says verbally as a text message they can respond to in the same way. The caller then hears what the user wrote from an AI-generated voice. The new Custom Voice Creator feature enhances that interaction by enabling users to submit voice samples that Bixby can employ to train a synthetic voice model mimicking them. Someone calling the user would then hear that voice reading out the message sent with Text Call. For security, the AI training requires the user to say specific sentences, with the initial audio recordings deleted once the setup is over. In addition, all of the text-to-voice processing happens on compatible devices, including the Galaxy S23, with nothing sent to the cloud. The custom voice creation Bixby feature is limited to Korean right now, but English and other devices are on the way, according to Samsung.
“When Samsung first launched Bixby as a voice assistant, it was part of a vision to create a human-to-machine interface that makes life easier and can advance over time with Samsung Galaxy innovation,” Samsung executive vice president YoungJip Kim, said. “Today, we introduce updates to build a more intelligent interface that is proactive and adaptive, giving people greater control over their mobile experience.”