Skip to main content
Version: 2026.07.0

Speaker Identification Changelog

For changes released before Speaker Identification 1.0.0, see the Voiceprint Extraction changelog and Voiceprint Comparison changelog.

1.0.0 (2026-10-12)​

Added​

  • [VOX-3091] - Initial release of speaker identification. Implemented endpoints are voiceprint extraction, voiceprint comparison, voiceprint merging and voiceprint conversion.
  • [PLAT-1714] Support for speaker identification model v5.5.0

Changed​

  • [PLAT-1711] All options with default values are now logged with their values, automatically detected value is logged for --num_threads_per_instance instead of 0

Fixed​

  • [PLAT-1714] GPU memory consumption no longer grows over the lifetime of the microservice as the load changes. The issue was fixed by a combination of a code fix in the microservice and a fix in the model (v5.5.0). Model v5.4.0 is still supported, but it is now slightly slower. Using the new model v5.5.0 is preferred, it is much faster.