User guide
Day-to-day use: importing a take, choosing what to transcribe, tracking progress, searching across mics, reviewing a transcript, and exporting.
01App layout
The top menu bar has File (import, export, close tab), Edit (copy, select all), View (switch between Dashboard and Viewer), Window (open tabs), Help, and — only for admin accounts in shared mode — Admin (user management, see Administration). Below that sits a tab switcher for the two main screens:
- Dashboard — the processing queue: every job, its status, progress, and bulk controls. This is where imports land and where you decide what actually gets transcribed.
- Viewer — a tabbed reading/playback view of finished (or in-progress) transcripts, with an audio/video player synced to the text.
02Importing plain files
File > Import Files... opens a dialog with:
- Transcribe audio streams separately
- If a file has more than one audio track (e.g. a multi-channel recording), each track becomes its own independent sub-job with its own progress, rather than being mixed down into one.
- Diarize speakers
- Runs speaker attribution on the result — see How It Works — diarization for what this does differently on a single mixed track vs. separate tracks.
- Sound Report CSV (fallback)
- Optional. Maps a file name to a character name using the same CSV format as MXF import, for cases where the file name itself matches a row in that report.
- Downsample / Denoise
- Optional pre-processing. Leave at defaults unless you have a specific reason to change them.
Click Import and the job appears on the Dashboard immediately and starts processing according to the queue.
03Importing an MXF take
File > Import MXF Folder/ZIP... is the workflow for production sound delivered as MXF takes with one physical track per microphone. It's a two-stage flow, deliberately:
-
Point at the source
Choose a folder or a ZIP of MXF takes, and optionally a sound-report CSV (only needed for tracks whose character name isn't already embedded in the MXF's own tags).
-
Scan
Clicking Scan reads every take and every track's metadata and lists them all immediately — this step does not transcribe anything yet, it just inventories what's there, tagging each track
ISO(an isolated per-character mic) orMIX(a pre-mixed track). -
Select & transcribe
On the Dashboard, the scanned tracks sit in a pending list. Use the mic search (see below) to find specific people/characters, check the ones you actually want transcribed, and click Transcribe Selected. Nothing runs on the GPU for a track until it's explicitly selected this way.
This scan-then-select design exists because a take folder can contain far more mic tracks than anyone needs transcribed for a given purpose — scanning is cheap, transcription is the expensive step, so the app never commits GPU time to a track without an explicit choice.
04The Dashboard
At the top, live gauges show CPU, GPU, and memory load. Below that:
- Processing Queue — every job as a card, reorderable by drag (top runs first), each showing per-stream progress.
- Export All (Combined) — merges every finished job into one chronological script. See Combining several jobs.
- Pause All / Resume All / Cancel All — queue-wide controls.
- A per-card ✓ exclude toggle on each finished stream, to drop just that mic from the combined export without re-running anything — see below.
05Searching & bulk actions on mics
The mic search box filters by name across every job and pending track, with autocomplete suggestions as you type. Two matching modes, both usable at once:
| Plain text | Matches as a substring — plant matches every "Plant Mic," useful for a whole family of related names. |
|---|---|
| "Quoted text" | Matches exactly — "Hero" matches only a mic literally named Hero, not "Other Hero." |
| Comma-separated | Combine several terms as an OR — e.g. Anthony, "Hero", plant. |
Once filtered, four buttons act only on the matching tracks:
This is the fast path for "everyone named X across every take" without clicking through each job individually.
06The Viewer & player
Switch to Viewer (or View > Viewer) to read a transcript. The sidebar lists jobs; opening one adds a tab. Each tab shows the turn-based transcript — speaker, timestamp, text — and, for an audio/video source, a player bar with play/pause and a seek bar whose displayed time accounts for the source recording's own timecode offset, so what's on screen matches what a production's own timecode-based notes say.
07Exporting a transcript
File > Export Transcript... (or the export control on a finished job) offers three formats:
Running as the desktop app, export opens a native Save dialog so the file goes exactly where you choose; running in a browser against a shared instance, it downloads through the browser's own download flow instead.
08Combining several jobs into one script
Export All (Combined) on the Dashboard merges every finished job's transcript into a single chronological script — useful when a scene was captured as several separate takes/imports but should read as one continuous document. If a particular mic ended up included by accident (e.g. a boom/mix track that duplicates what isolated mics already captured), click its ✓ button on the job card to exclude that one stream from the combined export; this is reversible at any time and doesn't touch that job's own individual transcript or re-run any processing.
09Live microphone transcription
The Live tab (when enabled) records straight from a connected microphone instead of importing a file. Pick a model and language, press REC, and speak — text appears as pauses are detected and each segment finishes transcribing, with a running chunk count and a live input-level meter. Press the record button again to stop; the session behaves like any other finished job afterward (viewable, exportable).