`transcript_aiprep`

Prepares transcript text for TTS synthesis using Claude. Fixes grammar, expands acronyms, converts numbers and symbols to spoken form, and adds phonetic hints for ambiguous terms.

When to use

Run before elevenlabs_tts or heygen_speaker when the source transcript contains raw text that would sound unnatural when read aloud (e.g. 99%, $10M, CQRS, kubectl).

In the editor

Add it from the Library (Transcript group), or with + add step and a search for transcript_aiprep, between the step that produced the words and the step that will speak them.

Under Inputs, add source: a step input pointing at your script or transcript step, or a file input naming a text file you uploaded. target is a dropdown, and it matters: pick the voice service the audio is going to, because the two want text written slightly differently.

Nothing appears on the timeline. The step passes on a cleaned version of the words, which the voice step then reads aloud, so 99% is spoken as ninety-nine percent rather than as a symbol, and your product names survive intact. Wire this step, not the original script, into the voice step.

YAML Example

The source input accepts either a text file or a step output. To chain from write_script, wire it directly:

- id: write_script
  processor: write_script
  # ...

- id: prep_script
  processor: transcript_aiprep
  input:
    - id: source
      step: write_script     # write_script emits a Script; step inputs are accepted
  params:
    target: "elevenlabs"
  cache: true

To prep a static file instead, use file: script-text for the source input.

Inputs

ID	Source	Description
`source`	step or file	Transcript or script text. A `step:` reference works (e.g. `write_script`, which emits a `Script`, or a `transcript_aiprep`/analysis output); a `file:` reference reads the text file directly.

Params

Param	Type	Default	Description
`target`	select: `heygen`, `elevenlabs`	`"heygen"`	TTS target: `"heygen"` or `"elevenlabs"`. Adjusts the cleaning rules for each system's quirks.

Output

Analysis containing cleaned script text ready for TTS.

Notes

Uses Puffin Ship's built-in AI: no API key required.
The cleaning rules include: grammar fixes, brand name preservation (Redpanda → Redpanda, not "red panda"), acronym expansion where needed, numbers to words (10M → "ten million"), symbols to words (99% → "ninety-nine percent"), phonetic spelling for technical terms that TTS mispronounces.
Use cache: true: the output is deterministic for the same transcript.