v1.3.0 — 2025-03-01
Deepfake detection now included on all verification calls. AASIST-based deepfake detection is now enabled by default across all plans and all verification calls. Thedeepfake_detected field in webhook results and verification responses reflects the result of this check. No integration changes are required to receive the signal — if you are already handling webhook results, deepfake_detected will now always be present in the payload.
We recommend auditing your webhook handler to ensure you are evaluating deepfake_detected before your match score threshold logic. See the Deepfake Detection guide for the correct decision tree and application logic patterns.
New: device_fingerprint parameter on verify endpoint.
You can now optionally submit a device fingerprint alongside verification audio. When provided, Voxmind records the device association and makes it available via the Devices API. This enables device-pattern analytics — understanding whether a user is authenticating from a consistent set of devices or from an unusual new device. Submitting a fingerprint has no effect on the verification result itself; it is purely for your analytics and fraud tooling.
Improvement: inconclusive result now includes a quality_signal field.
When a verification returns result: inconclusive, the webhook payload now includes a quality_signal field with one of three values: low_volume, excessive_noise, or insufficient_duration. This gives your application enough information to show the user a targeted prompt — “speak closer to your microphone” versus “find a quieter location” — rather than a generic retry message. This is a non-breaking addition.
v1.2.1 — 2025-02-15
Bug fix: match score precision in webhook payloads. In some edge cases,match_score was being returned as an integer (1) rather than a float (1.0) when the score reached the maximum value. This caused type errors in strictly typed applications. The field now consistently returns a float across all result values. This was a serialisation bug only — the underlying score calculation was unaffected.
v1.2.0 — 2025-02-01
New: Predefined Texts API. You can now manage your organisation’s voice prompt templates via the API rather than only through the dashboard. This enables programmatic management of multi-language prompt libraries — useful for organisations operating across multiple locales. See the Predefined Texts API reference for the full endpoint documentation. New: Organisation Settings API. Thealert_on_failure setting is now configurable via API in addition to the dashboard. When enabled, Voxmind sends an alert to your organisation’s contact email when verification failure rates exceed baseline thresholds — a useful early-warning signal for both fraud events and integration issues. See Update Settings.
Improvement: Enrollment now returns estimated voiceprint quality.
The enrollment webhook payload now includes a voiceprint_quality field with a value of good, acceptable, or marginal. A marginal quality score indicates the enrollment audio was processed successfully but the voiceprint has lower confidence than ideal — typically because the audio was short, noisy, or captured in a degraded telephony environment. For marginal enrollments, we recommend prompting the user to re-enroll under better conditions before relying on the voiceprint for high-value authentication. See Enrollment Best Practices.
v1.1.0 — 2025-01-15
New: Multi-language support expanded to 17 languages. The XLSR-300M phoneme extraction model now supports 17 language configurations, with improved performance across Arabic, Mandarin, and Brazilian Portuguese compared to the previous version. Because Voxmind is text-independent, these improvements apply to both enrollment and verification without any integration changes. See Language Support for the full list of supported ISO language codes. New: Filtering, sorting, and pagination on all list endpoints. AllGET endpoints that return collections now support filter, sort, per_page, and page query parameters. See Filtering, Sorting & Pagination for the complete syntax reference.
Improvement: Reduced verification latency.
Median verification latency has been reduced from 2.8 seconds to 1.9 seconds through pipeline optimisation. The latency reduction applies to all verification calls and does not require any integration changes. The maximum SLA remains 5 seconds.