VoxScan – Real-Time Voice Deepfake Detection for Trustworthy AI & Digital Communications

VoxScan is CandyVoice’s real-time voice deepfake detection technology, designed to identify synthetic, cloned and manipulated speech and strengthen trust in voice-based digital interactions.

The technology analyses acoustic, statistical and latent characteristics of speech to detect artefacts generated by TTS and speech-to-speech (S2S) systems. It operates without requiring a reference voice model or prior voice signature, making it suitable for open and distributed environments where speakers are not known in advance.

VoxScan is language-independent, GPU-free and designed for real-time edge deployment. It can run on embedded or server architectures and be integrated into communication systems, digital platforms and secure infrastructures. Its low computational footprint enables deployment close to the data source, reducing dependence on cloud processing.

For Horizon Europe collaborative projects, CandyVoice can contribute VoxScan as a technological building block for trustworthy AI, synthetic-content detection, voice authenticity and secure human-machine interactions, particularly in distributed and edge computing environments.

Potential research and integration activities include:

• detection and analysis of AI-generated or manipulated voice content;

• voice authenticity and trust mechanisms for AI-enabled services;

• edge-based detection within distributed computing architectures;

• integration with data, communication and AI platforms;

• evaluation of robustness across languages, environments and emerging generative voice technologies.

CandyVoice is interested in joining European consortia where VoxScan can be integrated, validated and further developed as part of larger trustworthy AI and distributed digital ecosystems, notably in the context of Horizon Europe Cluster 4.