Detection Layer
(VAAD) Voice AI Activity Detection Layer
Last updated
(VAAD) Voice AI Activity Detection Layer
Last updated
Secure your Communication with our Detection Layer
Clone X features an advanced Voice AI Activity Detection (VAAD) layer with an exceptional 99.9% recall, ensuring that every instance of AI activity—be it voice generation or alteration—is accurately detected and reported to the user. Both incoming and outgoing calls are monitored in real-time, delivering a seamless experience while maintaining the highest level of security for our customers. When AI activity is flagged during a call, it is automatically added to a blacklist, and the user is promptly alerted that the call may be fraudulent. This empowers customers to exercise heightened caution, reinforcing trust and safety during ongoing calls.
At the heart of the VAAD system lies our detection model, which leverages advanced deep learning technologies to identify patterns in voice data indicative of AI activity. This sophisticated model is trained on an extensive dataset comprising over a million audio samples, balanced across three distinct labels: AI-generated, AI-altered, and Human.
The dataset is strategically divided into three subsets to optimize model performance: 80% for training, 15% for testing, and 5% for validation. This distribution ensures the model avoids overfitting, enabling it to reliably identify these labels across diverse contexts. As a result, the model generalizes its knowledge effectively to multiple domains and scenarios.
For scenarios requiring enhanced precision, the model can be fine-tuned to specific domains, organizations, or even individual profiles. This flexibility ensures an unparalleled level of security and adaptability, empowering the VAAD system to meet the most stringent detection requirements.
As detailed above, the model is seamlessly integrated into a sophisticated pipeline designed to enable real-time inference during ongoing calls. This pipeline intelligently splits the incoming voice stream: one path delivers the voice directly to the customer, ensuring an uninterrupted experience, while the other routes it to the analysis server. The analyzer server evaluates the voice stream for AI activity and can, if necessary, push notifications or warnings back to the customer in real-time. This architecture ensures both an exceptional user experience and robust security, providing instant alerts when AI activity is detected.
When a number is flagged, it is immediately blacklisted, and we strongly advise against returning the call.
Automatic Ticket Creation: A ticket is instantly generated, notifying your organization, Clone X, and, if configured, the relevant authorities.
Comprehensive Filtering: A robust filtering system continuously scans blacklisted numbers for indicators of spam, fraud, or harassment, ensuring ongoing vigilance.
Real-Time Detection and Protection: All detection processes and protective actions occur during the call, delivering real-time security without disrupting the user experience.
For incoming calls, the system proactively checks the caller ID against the blacklist before connecting the call. If a number is flagged as spam, fraudulent, or harassing, the call is immediately blocked, preventing it from reaching the user and safeguarding your organization from potential threats.
By integrating real-time protection and advanced detection mechanisms, Clone X ensures your communications remain secure, reliable, and resilient against AI-driven and other malicious activities.
