FrameQuery 1.2.9: Automatic Audio Sync and Background Mode

FrameQuery 1.2.9 syncs external recordings to camera clips on its own, keeps watching your footage folders from the system tray, and processes BRAW, R3D and Nikon N-RAW through the camera SDKs. It also moves to a smaller face recognition model and stores its caches in less disk space. Nothing needs re-indexing.

What's new in 1.2.9

• Automatic audio sync: recorder audio attaches to camera clips once the timing checks pass. Uncertain matches wait for review.
• Microphone mixes: play aligned recordings together, solo a source and set levels before processing.
• Other angles: camera clips synced to the same recording list each other and open at the same moment.
• Background mode: keep running in the tray, start when you sign in, and keep the computer awake through a batch.
• Alerts: one inbox for finished batches, failed videos, low processing hours and folders that can't be reached.
• First-run setup: choose what you film and which folders to watch, and the defaults follow from that.
• RAW processing: BRAW, R3D and Nikon N-RAW are processed with the same camera SDKs used for playback.
• RAW playback: fixed an issue with BRAW freezing after a scrub preview, and 12-bit ProRes 4444 in MXF now plays.
• Smaller face model: a 39 MB download in place of 191 MB. Your named people stay.
• Voices across videos: fixed an issue with recognising a named speaker in other videos.
• People tab: loads in pages on large libraries, with drag-to-assign and sorting.
• Less disk: sampled frames stored as H.264, smaller posters, and smaller frames for scenes without faces.

How do you sync an external recorder to camera clips automatically?

Add the folder with your recorder files as a source next to the camera footage. FrameQuery indexes the recordings on this computer and checks each camera clip against them. When source, channel and timing agree across the clip, the recording is attached, and the Audio Sync panel on the video page shows it with Undo automatic match.

Automatic audio matching is on by default; the switch is under Settings, Media Scanning. Camera clips are only read for matching once there are recordings to match against. A match with uncertain timing waits for review, and a choice you make yourself always wins.

To attach a take by hand, open Find or attach recording, mark the same clap or word in both waveforms, check the alignment and attach it. Playback uses that recording, and processing sends it for transcription in place of the camera audio.

When several recordings pass the checks, Edit microphone mix lays them out on the video's timeline to solo, drop or level. Check mix samples the start, middle and end for delayed duplicates and signals that cancel out. It won't catch room echo or bleed, so listen first.

Other angles lists the camera clips synced to the same recording and the part of this clip each one overlaps. Click one and it opens at the same moment. Two cameras with no recording between them aren't linked yet.

On a 52-hour test library the audio index went from 555 MiB to 258 MiB, and peak memory while indexing roughly halved. See dual-system audio sync for the camera-and-recorder workflow and shoot to selects for multicam interviews for where Other angles fits.

Leave it running: watched folders, the tray and alerts

Turn on Keep running when the window is closed under Settings, Running in the background, and closing the window leaves FrameQuery in the system tray, watching your folders and processing new footage. Start when I sign in to my computer opens it straight to the tray. Keep the computer awake while processing is on by default; the screen can still turn off.

Watched folders get a catch-up scan at launch and again when a drive is plugged back in. Network shares are rescanned every 15 minutes and cloud folders every 30. A new file must stop changing before it is imported, so a card that is still copying isn't processed half-written. Jobs that were queued or running when FrameQuery closed resume on the next launch.

The bell in the header collects finished batches, failed videos, low hours and cloud folders that need reconnecting, and shows them as system notifications while FrameQuery is in the background. Pause 1h holds automatic scans and processing for an hour.

New installs start with a short setup: what you film, which folders to watch, whether to keep running. Automatic processing uses your processing hours and stops with an alert when they run out. The guide to indexing new footage automatically walks through an overnight setup.

Process and play RAW footage from the card

Processing BRAW, R3D (including takes split across files) and Nikon N-RAW now uses the camera SDK decoders that playback already used. Split takes play, scrub and process as one recording.

We fixed an issue with BRAW pausing for about 30 seconds after a scrub preview; playback now resumes in milliseconds. On our test machine, BRAW decode at the 95th percentile dropped from 62 ms to 20 ms per frame, and the ProRes RAW player peaks at 285 MiB of memory instead of 482 MiB. Rapid scrubbing drops stale frames, and BRAW and R3D audio starts at the seek point. We also fixed an issue with 12-bit ProRes 4444 and 4444 XQ in MXF, as recorded by ALEXA cameras, which now plays.

Processing scales frames on the GPU when an NVIDIA card is available, and 10-bit and 4:2:2 footage decodes on the GPU instead of falling back to the CPU. We fixed an issue with sideways scene frames from portrait phone clips decoded on the GPU. Clips processed from 1.2.9 on are upright, and earlier ones keep their frames. We tested the GPU changes on Windows with an NVIDIA card and wrote up the details.

Faces, voices and the People tab

1.2.9 replaces the face models with YuNet for detection and SFace for recognition. The download is 39 MB instead of 191 MB. On our test frames, face detection on the CPU went from 64 ms to 6 ms per frame (median).

After the update, FrameQuery rebuilds each person's face references from frames it already saved, going back to the original video when it has to. Names and people stay as they are, and the People tab shows how many people are done. New face recognition waits for the rebuild, then the old model and its data are deleted.

We fixed an issue with voice matching across videos, so a voice you name in one video is recognised in the next. Each named person now has one voice profile, and speakers in new videos are matched against it.

The People tab loads in pages, so libraries with thousands of faces stay responsive. Drag a face onto a person, sort by name or newest, or select several faces and assign them at once. Likely duplicate people lists pairs that may be the same person.

Face and voice matching runs on this computer, and the templates stay on it. The people search guide covers naming and missing matches.

What's fixed

• Fixed an issue with AI search inside a project, or with a tag or metadata filter, returning few or no results.
• Fixed an issue with the tc: timecode filter. It now reads start timecode from ffprobe, ARRI, RED and BRAW metadata.
• Fixed an issue with poster size: posters averaged about 340 KB, and up to 30 MB for 8K RAW. Existing posters are re-encoded in the background.
• Fixed an issue with saving sources resetting a folder's auto-analyze and archive mode settings.
• Fixed an issue with scans losing found files when you left the Library page before they finished.
• Fixed an issue with the same file appearing in the library twice. Duplicates are merged after the update.
• Fixed an issue with typing in search stalling other parts of the app. Filename search on a 20,000-video test library also went from 365 ms to 88 ms (95th percentile).
• Fixed an issue with the scrub preview cache growing past its 2 GB limit.
• Fixed an issue with videos failing when a share was unreachable, the data drive was full or FFmpeg was missing. They now wait and retry.
• Missing scene frames are rebuilt from the source in the background.
• German, Spanish and French translations now cover the whole app.
• Error reports, sent only with analytics consent, are kept on disk until the next launch. They contain no file names or error messages.

Questions

Why is FrameQuery busy after updating?

It is running one-time conversions: the face model rebuild, re-encoding sampled frames and posters, shrinking frames for scenes without faces, merging duplicates, rebuilding the audio index and upgrading the search indexes. They run one at a time at low priority, and pause while you play a video, while FrameQuery processes new footage and while a laptop runs on battery. The Jobs tab shows Finishing the update with each one's progress. Disk use rises during the search index upgrade and drops again once the old copy is deleted. Single-frame H.264 vs JPEG has the measurements behind the frame conversion.

Can two computers open the same library folder on a NAS?

No. A library folder on a NAS opens on one computer at a time. A second computer now shows Library already open and closes, where 1.2.8 opened it read-only.

Can I go back to 1.2.8?

Not cleanly. Once the conversions have run, 1.2.8 can't read the new thumbnail folders or search index tables, so it shows blank scene thumbnails and empty AI results.

How to update

Install the update through FrameQuery, or use the download page for Windows and macOS on Apple Silicon. Existing videos, projects and people carry over, and the conversions above run on their own.

Two things behave differently on the first launch after updating:

• Clips that arrived in an auto-analyzed folder while FrameQuery was closed are listed but not processed automatically. An alert for each folder offers to process them.
• Keep the drives with your footage connected while the face rebuild runs. People whose videos are offline keep their names, but aren't recognised in new footage until you reconnect the drive and use Rescan project in the People tab.