AI-powered redaction for video, images, and audio

New Features

• SRT export for speech transcripts
  In addition to existing export formats (raw, JSON, and others), transcripts can now be exported as SubRip Subtitle (SRT). When an SRT file is placed alongside the exported video with the same base name (for example, out.mp4 and out.srt), compatible video players can overlay the detected speech as subtitles.

Updates

• GPU support for speech detection
  Speech detection now uses the same GPUs available for vision processing, and can be up to 10× faster than CPU.
• Improved speech detection
  Speech detection is more accurate, especially in difficult conditions such as very quiet or noisy audio.
• Improved speaker detection
  Speaker separation has been fine-tuned with new sample material and is more reliable in typical dialogue scenes.
• Create speakers in speech-only transcriptions
  When only speech was detected, you can now select text and create a speaker from the context menu. The selected text becomes the secondary speaker.
• Object list follows video seeking
  When seeking to a new timestamp, the object list centers on that time and shows the closest matching objects. This reduces unnecessary seeking among objects in the current scene.

Bug Fixes

• Speech UI: overly long text segments for continuous speech
  Timestamps are now forced at least every 20 seconds to improve readability.
• Mosaic at highest intensity rendered as a solid color
  Highest-intensity mosaic is now coarse but still shows distinct fragments.
• Failed renders with no output (zero-render cases)
  Renders could fail with no output when the destination was locked or did not exist. Workarounds are now in place to handle these cases.
• Videos remained in use after returning to the list view
  Videos are now released immediately for reopening; the original close work continues in the background.
• License problems reported as internal errors
  Expired or invalid licenses no longer surface as generic internal errors on actions such as export. Users now see a clear message describing the license issue.

‍

Bug Fixes

• Empty Video List after License Exchange (Desktop)
  Desktop app now refreshes the video list after license reactivation, so videos remain visible without restarting.
• Speaker Labeling
  Fixed speech detection speaker labels so numbering is contiguous and always starts with 'Speaker #1' based on first appearance.

‍

New Features

• Speech Transcription Improved
  Introduced a new engine for which the detection and accuracy of speech have been vastly improved. Installers and Docker images have shrunken in size, the english language is onboard, other languages can be downloaded after installation.
• Speaker Diarization (beta)
  Using speech detection one can also now attempt to detect who spoke. Speakers are identified, can be labeled, corrected and merged. This allows selection and redaction of certain persons. This is our first best effort release, fine tuning and adjusting it over the next releases will be ongoing.
• New Speech UI
  • New editor-like overall look and feel.
  • Maximized view, alternative to the side bar one.
  • Enhanced text search and selection (words, regexes, speakers, inversion, ...)
  • Export in multiple text formats (raw, JSON, ...)
  • Speaker identifier management (reassignment, relabeling)
  • Display options (colors, transparency, word presentation, ...)
  • Live animation of currently played words.
• Work Time Tracking
  Redactor can record how long editors actively spend on each session, with a start/stop clock and a live HH:MM:SS readout in the editor and a running total in the server video list. Administrators and supervisors can reset a video's recorded time. The actual time data is exported in the audit logs of a session.

Updates

• Richer Audit Archive
  The "Acquire Audit Data" archive now includes an info.json manifest describing the archive's origin (host, network addresses, and user) and, for each session, a SHA-256 checksum of its audit log together with the video's metadata. The creation and inclusion of history.gz has been removed due to performance issues (needs to be revisited).
• Smoother Editing of Long Transcripts
  The speech editor stays responsive on very long transcripts (an hour of audio and thousands of words), removing the lag that could appear when opening the panel or during playback.
• Smoother Waveform Zoom
  Zooming the audio waveform with the mouse wheel or a pinch gesture now redraws smoothly instead of stuttering during rapid zooming.
• Consistent Playback-Speed Picker
  The video playback-speed control now looks and behaves the same across browsers and the desktop app.
• Live Track Recordings Recovered After Disconnect
  A manual Live Track recording is no longer lost if the connection to the server drops before you click Apply. In-progress recordings are now staged periodically and committed automatically if the connection is lost. (The desktop version was not affected.)
• Session Preservation For Missing Video Files
  If video files are missing (most common case a different mount for the videos folder) sessions will no longer be removed but all of their data preserved. On the next start a rediscovery attempt will be made and if video files are back their associated sessions be restore.
• Corrupted Data Healing Improved
  Measurements have been put in place to correct corrupted redaction (e.g. due to bugs or other circumstances).

Bug Fixes

• License Availability Count
  A license with a fixed expiry date could incorrectly report zero available seats. Seat availability is now calculated correctly.
• Empty Context Menu in Videos List
  In the video, when selecting multiple files, the not selected entries showed an empty context menu stub.
• Admin Storage Shows Zero Bytes
  In some deployments the admin showed zero bytes storage usage, this has now been corrected.

‍