Skip to main content

Version 2026.07.0

· 5 min read

General changes​

Features​

  • Added nvtop for interactive GPU monitoring in the appliance console
  • Various improvements of the configure-speech-platform.sh configuration script

Fixes​

  • Manual (static) network configuration is preserved across VM restarts again. A regression introduced in 2026.05.1 — the Rocky Linux 10.1 upgrade shipped a newer cloud-init that re-rendered and overwrote the network interface on every boot.

Technology changes​

Features​

  • Grapheme to Phoneme Conversion and Phoneme Recognition have been added

  • Deepfake Detection uses a new version of the Generic model:

    • Covers new types of spoofing attacks
    • Has improved resilience to codec artifacts, noise, and other common degradations
    • Requires Deepfake Detection service version 4.2.0 or later

  • Support for Albanian has been added to Enhanced Speech to Text Built on Whisper via the large-v3-turbo-sq-v2 model

  • Speech Translation built on Whisper now supports translation to Korean

Fixes​

  • Enhanced Speech to Text Built on Whisper
    • Fixed issue where low-quality audio could trigger higher level of hallucinations or even completely missing transcription
    • Fixed a bug where Voice Activity Detection (pre-processing step of Speech to Text) ignored the GPU selection and always ran on the first available GPU in multi-GPU systems
    • Fixed an issue introduced in 2026.05.1 release, which caused high VRAM consumption that could eventually lead to a crash of the service
  • Keyword Spotting
    • Fixed a bug where single-grapheme keywords without explicit pronunciation didn't produce any output
    • Fixed a bug where pronunciation source was incorrect for keywords consisting of multiple words separated by +
  • Many technologies that process audio use a new, significantly faster resampler library which fixes slowdowns caused by a previous resampler change
    • As a consequence, these technologies can produce negligibly different results than the previous version
  • Language Identification, Gender Identification, Speaker Diarization, Referential Deepfake Detection, Emotion Recognition, Voiceprint Extraction, and Deepfake Detection have various internal changes and fixes that can change the results slightly

REST API changes​

Features​

Fixes​

  • Denoiser fixed a bug where the time range query parameters ware ignored

Documentation​

  • The REST API introduction contains instructions on robust handling of error messages
  • All endpoints now document the 500 Internal Server Error response, including its error format

GUI (web application) changes​

Features​

Settings menu​
  • In the Microservices tab of the System configuration section, the Media Conversion service has been renamed to Media Processing
Technologies​
  • The UI has received a major facelift across the majority of technologies (except Speaker Identification) with the goal of unifying the UI, enhancing responsiveness (better display on mobile devices), and improving user friendliness
  • The file table has new columns with result summaries for the given technology and with information about the number of channels
  • With technologies that allow input parameter modification (e.g., specifying language subsets in Language Identification), the "Input parameters" component has moved to the Results section
  • The audio player has moved to the bottom of the screen
  • On mobile devices, the detailed results are displayed by selecting a file from the file table
  • Technologies can be switched easily in the page header
Included Components
  • Age Estimation 2.1.0
  • Audio Manipulation Detection 2.1.0
  • Audio Quality Estimation 2.1.0
  • Deepfake Detection 4.3.0
  • Referential Deepfake Detection 2.1.0
  • Denoiser 2.1.0
  • Emotion Recognition 2.1.0
  • Enhanced Speech to Text Built on Whisper 2.5.0
  • Gender Identification 2.1.0
  • Grapheme to Phoneme Conversion 1.1.0
  • Keyword Spotting 2.2.0
  • Language Identification 2.1.0
  • Media Processing 2.0.0
  • Phoneme Recognition 1.1.0
  • Replay Attack Detection 2.2.0
  • Speaker Diarization 2.1.0
  • Speech to Text 2.2.0
  • Text Translation 2.1.0
  • Time Analysis 2.1.0
  • Transcription Normalization 2.1.0
  • Voice Activity Detection 2.2.0
  • Voiceprint Comparison 2.1.0
  • Voiceprint Extraction 2.1.0