A newly issued patent for synthetic voice detection will be built into Daon’s call center fraud protection platform to secure voice biometric authentication.
“Methods And Systems For Enhancing The Detection Of Synthetic Voice Data” has been approved by the U.S. Patent and Trademark Office. It describes a conversion of monophonic voice data into stereophonic voice data, which is decomposed “into a mid-signal and a side signal” with a machine learning model. The signals are then analyzed for artifacts that indicate synthetic generation.
The patent names Daon President and CPO Ralph Rodriguez as a co-inventor, along with Olena and Davyd Mizynchuk. The company has around 50 researchers working on innovations in biometrics, AI, deepfake detection and related areas.
Rodriguez tells Biometric Update in an email that the patented technology will be implemented in Daon’s xDeTECH voice fraud detection software to help identify synthetic and manipulated voices. xDeTECH, formerly known as xSentinel, is part of Daon’s AI.X suite of deepfake security solutions, and is included as a standard feature with the company’s xVoice offerings.
The same technology could also be used to detect deepfakes on social media, he says.
Like authentication with face biometrics, Rodriguez believes voice authentication systems are vulnerable without layers protecting against sophisticated spoofs.
“Without synthetic voice detection in place, it is likely that some deepfake voices produced by tools, such as ElevenLabs, might pass through biometric voice authentication,” he writes in the email. “While biometric algorithms primarily assess biometric patterns, our patented deepfake voice system called xDeTECH, is designed to identify anomalies associated with synthetic voices, thus making it a critical layer to intercept deepfakes that would otherwise evade standard biometric checks.”
Replay detection and voice quality assurance do not provide adequate protection against deepfakes, he says.
xDeTECH can also be integrated with major telephony platforms from Genesys, Amazon Connect, Cisco, Avaya and others. Rodriguez suggests other applications of the synthetic voice detection methods specified in the patent include voice biometric authentication for workforce IT help desks, as well as protecting telecom, healthcare or government systems.
Splitting signals into stereo
Many deepfake and synthetic speech generation models do not accurately replicate the intricate details of speech, Rodriguez says, resulting in anomalies that are detectable, but not necessarily easy to isolate in the samples call centers and others using the voice channel have to work with.
This is why the patented method starts with splitting the audio data.
“Stereophonic processing enables a richer analysis by splitting audio data into distinct channels, allowing the system to detect subtle frequency anomalies, phase shifts, and inconsistencies in harmonic structures that may not be evident in monophonic data,” Rodriguez says.
The USPTO has also informed Rodriguez that it will issue approval for his patent application for “Methods and Systems for Enhancing Detection of Fraudulent Data” by the end of the year.
Article Topics
biometric authentication | call centers | Daon | deepfake detection | patents | synthetic voice | voice authentication | voice biometrics
