Welcome to xDeTECH
What is xDeTECH?
xDeTECH, Daon's Detection TECHnology for voice replay, synthetic audio, and cloning is a service provided by Daon designed to identify instances of cloned and voice replays during telephone calls. xDeTECH components sit at the point of entry for calls, providing an early and accurate signal as to whether the audio has been synthetically generated or audio recordings played back into the call.
xDeTECH Components
Voice Clone Detector (VCDx)
VCDx is a REST-based microservice that will take audio samples and perform a quality assessment, detect audio playback/audio replay, and detect synthetic and cloned voice. This service analyses short audio samples belonging to a call and returns a result indicating whether a replay or clone has been detected. This service can be called directly by an external service which can create chunks of audio from a call itself, or indirectly via an audio streaming service and the xDeTECH Voice Gateway (VGSx).
VCDx results include:
- Quality Assessment - Audio classification and quality estimation to decide whether the sample provided is of sufficient quality to perform an analysis. If quality is not sufficient, replay and clone detection will not proceed.
- Replay Detection - Audio analysis for replay artifact detection to decide whether to reject or pass-through
- Clone Detection - Speech analysis to detect artifacts generated by the current 23 known Voice Cloning (VC) systems
Voice Gateway Service (VGSx)
This service acts as a gateway between the a call audio streaming service and the VCDx. The VGSx streams call audio and splits it into chunks which are sent to the VCDx for analysis. The status of the current analysis of a call (clone detected or not) can be retrieved from VGSx. The VGSx is designed to use OrkAudio as the streaming service which supplies real-time call audio.
SIPREC Server (OrkAudio)
The diagram below shows at a high-level how the xDeTECH VGSx integrates with the OrkAudio SIPREC Server. The SIPREC Server connects to the VGSx over HTTP(S) and announces the start and end of voice calls it receives. The VGSx will, on receiving an action to start analyzing a voice call, shown here coming from an Agent Workstation as an example, will request the audio stream from the SIPREC Server via HTTP(S) and process it.
xDeTECH Voice Gateway Tester (VGSx Tester)
This is a simple streaming service which accepts files as audio and streams the file audio to the xDeTECH Voice Gateway. This component is intended for test purposes and is not intended for production use.
xDeTECH Web App
Provides a user interface for testing the xDeTECH service. The web app allows files to be uploaded and streamed as if they were calls to the xDeTECH Voice Clone Detector and shows the results of call analysis. This component is intended for test purposes and is not intended for production use.
The on-premise service needs access to Daon TrustX cloud services for license management and to publish audio analysis statistics.
TrustX Connectivity
The xDeTECH VCDx service requires connectivity to the TrustX authentication & authorization service (ARTHR) to issue JWTs and connectivity to the TrustX Sentinel service to ping and to validate JWTs.
Your on-prem VCDx service must be able to connect to the following URLs:
Details on configuring <tenant> and <sky region> are found in the Initial Configuration section of this guide.
Note: Access to the https://system.<sky region>.trustx.com/api/sentinel/actuator/health URL is restricted to IP address(es) and the Daon Support team may ask for your IP address range.