Skip to content

Transcribe Audio and Video Files

Goal

Turn an existing audio or video file into a readable transcript, optional subtitles, and an optional summary. Use the same setup for a whole folder when you have more files to process.

Why This Pattern Exists

Batch transcription solves a different problem from live dictation:

  • Inputs already exist on disk.
  • You need queue-level throughput and status visibility.
  • Outputs must be reproducible across many files.

The file-transcription pipeline supports queue states, per-item output folders, optional segmentation chunking, subtitle export, and optional LLM outputs.

Start with One File

  1. Open Transcribe File(s)... and choose Add Files..., or drag in an audio or video file.
  2. Choose an output directory and enable TXT. Add SRT for an editor/player or WebVTT for a web player if you need captions.
  3. Leave LLM output off for a transcript, or select a tested summary processor and enable .llm.txt for a companion summary.
  4. Click Transcribe and wait for the item to show Completed.
  5. Use its Open Folder action to read the transcript and check any subtitles against the source media.

For a larger queue, use the preset and batch sequence below after checking one representative result.

Preset: Batch Standard

Setting Where Value Why
Output directory Transcribe File(s) Settings -> Output ~/Documents/OpenScreech File Transcriptions or project path Keeps batch artifacts isolated from live recordings.
Chunking Transcribe File(s) Settings -> Audio Quality On Segments long transcripts into useful sections.
Chunk (s) Transcribe File(s) Settings -> Audio Quality 120 Balanced section size for review workflows.
Word-level timings Transcribe File(s) Settings -> Output On Supports subtitle and timeline tooling.
TXT output Transcribe File(s) Settings -> Output On Primary readable transcript output.
SRT output Transcribe File(s) Settings -> Output On Subtitle compatibility for editors/players.
WebVTT output Transcribe File(s) Settings -> Output On Web workflow compatibility.
TTML output Transcribe File(s) Settings -> Output Off unless required Avoid extra files unless needed.
SMPTE-TT output Transcribe File(s) Settings -> Output Off unless required Enable only for broadcast requirements.
Cue split mode Transcribe File(s) Settings -> Output Sentence Better human readability in subtitle cues.
LLM post-processor Transcribe File(s) Settings -> LLM Post-Processor A dedicated batch processor Keeps analysis behavior explicit and reusable across batches.
LLM Post-Processor Output (.llm.txt) Transcribe File(s) Settings -> Output Off by default Saves the selected post-processor's result without changing the original transcript.
Save LLM Request Payload Transcribe File(s) Settings -> LLM Off unless troubleshooting Reduces unnecessary artifacts.

Preset: Research/Analysis Variant

Use this when you need post-processed summaries per file.

Setting Value
LLM Post-Processing On
LLM Request Payload On
LLM processor A tested analysis processor from the shared library
Processor timeout 60
Processor model Your larger analysis-capable model

Queue Operation Playbook

  1. Open Transcribe File(s)....
  2. Use Add Folder... for source directory.
  3. Confirm queue count and remove accidental inputs.
  4. Select the intended LLM processor, or leave LLM output disabled.
  5. Before a large LLM batch, use Edit Processor... and Test LLM Post-Processing... with representative input.
  6. Confirm the remaining settings preset.
  7. Click Transcribe.
  8. Watch item statuses (Queued, In progress, Completed, Failed).
  9. Use Open Folder per completed item for QA.

QA Checklist Per Item

  • Transcript file exists and has expected language/content.
  • Subtitle files open correctly in your target tool.
  • If LLM enabled: compare .llm.txt against source transcript for hallucination drift.

Tuning Guidance

  • If processing is too slow: disable LLM and reduce enabled artifact formats.
  • If sections are too large for reviewers: reduce Chunk (s).
  • If subtitle pacing is awkward: tune cue split mode and sentence/word thresholds.

Quick Tips

  • Test one representative file before starting a large queue, especially when LLM output is enabled. It validates segmentation, prompt quality, and the chosen artifact set at the cheapest point.
  • A shared post-processor can change between batches because it is a library reference. Duplicate it when a particular batch must remain tied to a stable prompt and endpoint configuration.

See Also

Visual Notes

  • TODO: add screenshot of queue with mixed statuses.
  • TODO: add screenshot of file transcription settings panel.
  • TODO: add GIF processing an entire folder and opening outputs.