Release Notes

xDeTECH 2.0

The xDeTECH 2.0 release provides improvements focused on providing a more accurate response earlier in a call. This is achieved by:

• Reducing the number of audio segments that are rejected due to quality checks. xDeTECH 2.0 now accumulates speech across multiple audio segments to reach minimum aggregated speech for a call sooner
• Processing results will be returned earlier in a call due to aggregating speech and a NO_ANOMALY_DETECTED decision will be returned if the first processed audio segment is found to be acceptable.
• New clone detection models and updated thresholds offer enhanced detection and lower BPCER

VCDx 2.0.0.0

Released: October 2025

Container Release Tags: sentinel-on-prem:vcdx-gpu-4373248

Algorithm Versions: DaonVoiceQC 2.0.1.0, DaonVoiceReplayTelephony 2.2.0.1, DaonCloneTelephony 2.0.0.1

What's New

• A new Voice Cloning Detection Algorithm has been introduced and includes the following improvements:

  • Enhanced presentation attack detection with improved accuracy and reduced BPCER. Presentation attack detection now offers robust protection against synthetic speech through expanded coverage of 150 voice cloning systems.
  • Optimized the decision-making process by reducing the number of audio chunks required, resulting in faster and more efficient evaluations.
  • Detection now triggers only after sufficient speech has been collected, ensuring more accurate and consistent results.
  • In xDeTECH 1.x, voice quality assessment was performed in parallel to clone and replay detection. With xDeTECH 2.0, voice quality assessment is performed ahead of the clone/replay detection and will lead to longer processing times.

• Cache changes:

  • To handle the aggregation of audio data, speech template data is persisted in cache until sufficient audio is collected and that speech is processed (or time expires). This results in higher Redis cache usage.

• A new property has been added in the VCDx qc.template.inference.duration

• Addressed CVEs:

  • High: CVE-2025-48988, CVE-2025-52999, CVE-2014-4043
  • Medium: CVE-2025-49125, CVE-2025-48924, CVE-2025-53864, CVE-2025-22227, CVE-2025-41234

• API changes:

  • Additional model and changes to model names has resulted in minor API changes in the API response. Please refer to the separate section: The VCDx API.

VGSx 2.0.0.0

Released: October 2025

Container Release Tags: sentinel-on-prem:vgsx-9e5f0ea

What's New

• Reduced the time that a call is kept in memory from 2 hours to 1 hour.
• Limited the number of raw audio samples per call to 500, and introduced a new property to manage this value: audio.sample.per.call.limit property.
• Updated the maximum number of attempts to 1000 to limit the number of audio samples sent to VCDx per call.
• Updated endpoint to track the total count and size of raw audio and speech samples currently in memory.

All-In-One Container Release

Released: October 2025

Container Release Tags: sentinel-on-prem:98b989d

Algorithm Versions: DaonVoiceQC 2.0.1.0, DaonVoiceReplayTelephony 2.2.0.1, DaonCloneTelephony 2.0.0.1

What's New

• Includes the above VCDx and VGSx 2.0.0.0 releases
• Numerous CVE fixes have been made since the previous All-In-One release.

Known Issues

Supported Platforms

VCDx is provided as a docker container with support for CPU and GPU deployments. Production systems are strongly encouraged to run on GPU. Daon validates and tests on Amazon AWS g4dn.xlarge instance types and recommends that instance size for production deployments. See Performance Testing and Scalability for more details.

VGSx is provided as a docker container with support for CPU deployments. 4 Cores and 16 GB RAM are recommended for VGSx.

Supported Browsers (Test Tool)

• Google Chrome
• Mozilla Firefox
• Safari