Skip to main content

Version 5.0.0

· 5 min read
  • Deepfake Detection technology is promoted from experimental beta to production-level status.
    In relation to this change, the technology is separated from Authenticity Verification into its own REST API endpoint /api/technology/deepfake-detection and also into its own tile in the GUI home page.
    The Authenticity Verification technology itself (with its Audio Manipulation Detection and Replay Attack subtechnologies) remains in the experimental status... and, being still experimental, is not billed.
    As a result of this separation, existing users of the affected technologies need to obtain models with updated license when updating the virtual appliance to this version.
  • Also, a new experimental Referential Deepfake Detection functionality is implemented.
    This functionality introduces a comparative approach that analyzes two recordings: a reference recording (known authentic recording of a speaker) and a questioned recording (which needs to be evaluated for authenticity). This functionality provides enhanced accuracy by leveraging the reference audio for comparison.
  • Speech Translation technology is enhanced with additional 16 target languages.
    In addition to English, which was already available in the previous releases, the audio transcription can now be outputted in Albanian, Arabic, Bengali, Chinese, Czech, French, German, Hebrew, Hindi, Malay, Persian, Portuguese (Portugal), Portuguese (Brazil), Russian, Slovak and Spanish language.
  • Audio Quality Estimation and Time Analysis technologies are now internally implemented as pure microservices (previously, the Phonexia Speech Engine was used internally). This is purely internal change, the API remains the same.
    As a result of this change, existing users of the affected technologies need to obtain models with updated license when updating the virtual appliance to this version.
  • Configuration and administration changes
    • Changes in limits configuration:
      • new maxVoiceprintListLength and maxVoiceprintMatrixSize limits for voiceprint comparison operations
      • new outputStorage.capacity limit for non-textual results temporary storage
      • renamed inputStorageSize to inputStorage.capacity
      • renamed internalStorageSize to internalStorage.capacity
      • renamed singleFileUploadSize to maxUploadFileSize
      • removed singleFileUploadTimeout limit as a result of implemented chunked transfer for files upload
    • Fixed the issue introduced in release 4.1.0, causing manual network configuration not being persisted across restarts of virtual appliance.
    • The 3rd party filebrowser component is updated to latest version, which fixes various issues like inability to overwrite existing file. This update also seems to fix the mysterious intermittent issue with extremely slow upload speeds when using VMware Workstation and using the GUI locally from the host computer.
      Also, as part of the update we also made a final fix for an issue which still occasionally extracted the uploaded licensed models ZIP file multiple times in a row.

REST API changes

  • Added experimental REST API endpoint /api/system/experimental/licenses for listing available licenses.
  • Implemented chunked transfer for files upload to REST API. This helps to keep the RAM consumption of the API component more stable and controllable, especially when using large media files.
  • Implemented possibility to cancel running tasks. So far it was possible to delete only finished task. Now it is possible to cancel also running tasks, in any stage of their life cycle.

GUI (web application) changes

  • Added button for discarding changes made during reordering tiles on the home page.
  • Added new About Results feature, available within each technology. It provides an overview of the technology's purpose, expected outcomes, and best-practices recommendations. It also offers hints on how to configure the technology, processing parameters, or audio settings to achieve optimal performance.
  • A new option has been added to General Settings, allowing to delete all custom settings and data at once, to fully restore the system to its original default state.
  • A new System Configuration section is now available in the settings menu as a beta feature. This feature allows advanced adjustment of the system environment. Access is secured by a password (available upon request) and intended for expert users.
    NOTE: Changes made here can significantly affect the behaviour of the Virtual Appliance and should only be made if you really know what you are doing.
  • Deepfake Detection is separated from Authenticity Verification and has its own tile on the home page.
    Additionally, for each uploaded suspected audio it is now possible to add an authentic audio as a reference, to improve the detection precision.
  • List of languages available for selection in Speech To Text and Speech Translation now reflects the actual languages enabled in the license. Previously, the Speech To text selection used a hardcoded list containing a full set of available languages, which could be confusing.
  • In relation to enhancing Speech Translation with additional target languages, the language selection dialog now allows selecting not only the source, but also the target language. When uploading files for processing, files with known non-audio/video extensions are now rejected from being uploaded. This prevents from various obscure errors when the system tries to process mistakenly uploaded non-media files.
  • Fixed issue when home page tile for Speech Translation was enabled always together with Speech To Text. The tile is now enabled according to actual availability of the Speech Translation technology in the license.
  • Fixed issue which prevented uploading files or changing processing parameters for uploaded files when some processing was in progress. The fix should also positively impact GUI performance due to various internal improvements and optimizations.

Included Components
  • Age Estimation 2.0.0
  • Audio Manipulation Detection 2.0.1
  • Audio Quality Estimation 2.0.1
  • Deepfake Detection 4.0.0
  • Referential Deepfake Detection 2.0.0
  • Denoiser 2.0.0
  • Emotion Recognition 2.0.0
  • Enhanced Speech to Text Built on Whisper 2.0.0
  • Gender Identification 2.0.0
  • Keyword Spotting 2.0.0
  • Language Identification 2.0.0
  • Replay Attack Detection 2.0.0
  • Speaker Diarization 2.0.0
  • Speech to Text Phonexia 6th Generation 3.62.0
  • Text Translation 2.0.0
  • Time Analysis of Speech 2.0.0
  • Voice Activity Detection 2.0.0
  • Voiceprint Comparison 2.0.0
  • Voiceprint Extraction 2.0.0