How to bulk transcribe a back catalogue
One file is a transcription problem; several hundred is a pipeline problem. The three things that decide cost and outcome are: reuse existing captions or transcripts instead of paying to redo them, handle failures per-item so one bad file doesn't sink the run, and decide the output format before you start rather than re-processing everything later.
Step by step
-
1
Inventory first
List every item with its duration and whether a transcript or captions already exist. Total duration is what your cost estimate is built from, and the already-have column is often a third of the catalogue.
-
2
Never pay twice for the same audio
Reuse published transcripts and existing captions. GPU transcription should only touch what has nothing usable.
-
3
Process per item, not per batch
A single corrupt file, a missing audio URL, or one timeout should skip that item and continue. Runs that abort on the first error turn a four-hour job into a four-day one.
-
4
Decide the output format up front
Notes, flashcards, an ebook, and a retrieval corpus want genuinely different shapes. Changing your mind afterwards usually means reprocessing, and reprocessing audio is the expensive part.
-
5
Spot-check quality early
Transcribe five representative items and read them before committing the whole catalogue. Accents, jargon, music beds, and crosstalk are much cheaper to discover on item five.
Doing it by hand
The whole process without any particular tool, if you'd rather build it yourself:
- Build a list of items with durations and existing-transcript status.
- Fetch published transcripts and captions where they exist.
- Queue the remainder for speech-to-text, one item at a time.
- Log per-item failures and retry them separately at the end.
- Convert the finished transcripts into your target format once.
Common questions
How long does a large catalogue take?
Transcription is roughly proportional to audio length divided by how many items you run in parallel. The wall-clock time is usually dominated by downloading audio, not by the model.
What does it cost?
Cost tracks GPU minutes, so it tracks total audio duration. The biggest single saving is not re-transcribing anything that already has a usable transcript.
Is speaker diarization worth enabling?
For interviews and panels it changes the output from a wall of text into something quotable. For a single narrator it adds cost and occasional mislabelling for very little gain.
The shortcut
SourceWeaver does every step above in one pass — reusing existing captions and transcripts, transcribing the rest on a GPU with optional speaker labels, and exporting in the format your destination tool actually wants. You can see real output before signing up.
Process content you own or created, content you're licensed or have permission to use, or openly-licensed and public-domain material. Whether any particular use is permitted depends on copyright law and the rights you hold — it's worth being sure before you start.