South Korean security firm Hancomwith has launched “Hancom AISRC Speaky,” an advanced security solution that combines voice biometric authentication with integrated deepfake detection. The AI-driven platform is designed to counter the rapidly growing threat of synthetic voice fraud, which has become a major concern for financial institutions, call centers, and other sectors that rely on voice for user verification.
The launch of Speaky comes as generative AI tools make it easier than ever for criminals to clone voices with startling accuracy, often from just a few seconds of sample audio. High-profile incidents, such as a recent scam in Hong Kong where deepfake technology was used to defraud a company of $25 million, have underscored the urgent need for more sophisticated defenses. Hancomwith, a subsidiary of the Hancom Group, is positioning Speaky as a next-generation weapon in this escalating arms race.
The Threat Landscape: The Rise of Voice Cloning
The proliferation of accessible and powerful AI has turned voice into a highly vulnerable attack vector. Unlike visual deepfakes, which can sometimes be identified by subtle artifacts, sophisticated audio fakes are notoriously difficult for the human ear to detect. This has enabled a new wave of social engineering attacks, from fraudulent wire transfer requests impersonating a CEO to scams targeting families with faked voices of loved ones in distress.
For businesses, the threat is particularly acute. Industries that have invested heavily in voice biometrics for authentication now face the risk that these systems can be bypassed by AI-generated spoofs. This reality is forcing a rapid evolution in the industry, moving beyond simple voiceprint matching to incorporate robust liveness detection—a core focus of the growing market for deepfake detection technologies. The severity of the threat has led major cloud providers to reconsider their voice biometrics offerings entirely, with AWS and Microsoft scaling back or discontinuing their voice authentication services amid mounting security concerns.
How ‘Speaky’ Works: Beyond Simple Voice Matching
Hancomwith’s Speaky platform is designed to function as a liveness detection layer for audio. Rather than simply matching a voiceprint, the system analyzes the nuanced characteristics of human speech to determine if the speaker is a live person or a synthetic creation.
According to the company, the AI model extracts and analyzes subtle vocal features such as a person’s unique intonation, cadence, and even regional dialect to distinguish between a genuine user and a spoofing attack. This approach requires a deep understanding of linguistic patterns, highlighting a critical trend in AI security: the need for hyper-localized models. Speaky was developed using a deep learning model trained on extensive Korean-language datasets and employs a proprietary “triple training strategy” to minimize data bias and improve its adaptability across diverse environments. This focus on language-specific data suggests that the most effective deepfake detection tools may be those tailored to specific regions and cultures, potentially giving local champions like Hancomwith an advantage over global competitors with more generalized models.
The technology is the first commercial product to emerge from a research and development partnership between Hancomwith and the AI Security Research Center (AISRC) at Seoul’s Soongsil University, a government-funded initiative focused on defending against advanced cyber threats.
A Broader Strategy of Public-Private Collaboration
Hancomwith’s efforts extend beyond commercial product development. The company is actively involved in a broader, national-level initiative to combat manipulated media. It is participating in an international joint study led by the Korean National Police Agency to develop a comprehensive “system for determining the authenticity of manipulated content.”
This collaboration signals a significant convergence of commercial identity technology and law enforcement forensics. The goal is not just to prevent a fraudulent transaction but to provide authorities with reliable tools to analyze digital media for evidentiary purposes in criminal investigations. Hancomwith plans to demonstrate its progress at the upcoming Korea Police World Expo (KPEX 2025), where it will unveil a multimodal detection system capable of analyzing voice, video, and text, along with an MLOps platform for managing the lifecycle of AI detection models.
This move into digital forensics could open up substantial new markets for biometrics companies, adapting their liveness detection and anti-spoofing technologies for use in legal and criminal justice systems. As the line between a failed login attempt and a piece of criminal evidence blurs, the technologies that underpin the voice biometrics industry are set to play an increasingly critical role in public safety and justice.
–
By Ali Nassar-Smith
