Redactor v7.3.2: GPU-accelerated speech and subtitle export

Redactor v7.3.2 moves speech detection onto the GPU (the same hardware already used for vision) for a substantial speed increase, and adds SubRip (SRT) subtitle export for transcripts. ‍

What's new

Speech detection now runs on the GPU

Speech detection uses the GPU alongside vision processing, so transcribing a recording takes a fraction of the time it did on CPU. On the desktop app, the GPU is detected automatically; server deployments enable it by passing GPUs into the container.

For teams working through long interview or bodycam recordings, this turns speech transcription from a wait into something closer to real time.

Export transcripts as SRT subtitles

Transcripts now export as SubRip (.srt) alongside the existing raw and JSON formats. Place the .srt next to the exported video with the same name, and any compatible player overlays the speech as subtitles, useful for review, disclosure, and accessibility.

Refined speaker detection ‍

Speaker separation has been retuned against new sample material and holds up more reliably in ordinary dialogue. You can also now create a speaker directly from selected text in a speech-only transcript.

Also in this release

  • The object list follows video seeking, jump to a timestamp and it centers on the objects in that scene

  • Long stretches of continuous speech now break at least every 20 seconds for readability

  • Highest-intensity mosaic keeps visible fragments instead of collapsing to a solid block

  • Renders no longer fail silently when the destination is locked or missing

  • Videos release immediately on return to the list, so they can be reopened without waiting

  • Expired or invalid licences now show a clear message instead of a generic error‍ ‍

Availability

Redactor v7.3.2 is available now for desktop and server. Existing customers can upgrade using the installers below; sessions and redaction data carry forward.

Server deployments running speech detection on the GPU need to pass GPUs into the container (--gpus all on Docker run, or the equivalent in your compose file). The desktop app detects the GPU automatically.

Full technical detail is in the v7.3.2 release notes.

Looking for the latest? Redactor may have shipped updates since this release, see the full release notes or the latest news.

Haris R.

Haris manages Product Marketing at Sighthound, where he leads GTM, content and positioning strategy. With a background in computer science and B2B SaaS, he bridges technical expertise with strategic marketing.

Next
Next

Sighthound Publishes Its Privacy-and-Evidence Approach to Responsible Vehicle Recognition