CastPolish v1.7.0: Polished PDF and Word Documents, Automatically

CastPolish v1.7.0: Polished PDF and Word Documents, Automatically

CastPolish has always turned your raw audio into clean, normalized sound and a searchable transcript. But if you wanted a readable document to actually hand to someone — a summary, the chapters, the full transcript laid out like an article — you still had to paste the text into ChatGPT or a word processor and tidy it up by hand. Not anymore.

As of v1.7.0, every job automatically produces two new files: a reader-friendly PDF and an editable Microsoft Word (.docx) document — right alongside the audio, captions, and interactive transcript you already get.

CastPolish Recent Jobs panel showing PDF and DOCX download links
Every finished job now offers PDF and DOCX downloads next to the existing formats.

What’s in the document

Both files share the same clean structure:

  • A title page with a short summary and keyword tags
  • A longer, multi-paragraph summary
  • The full transcript, organized into chapters with AI-generated headings and broken into readable, sentence-aware paragraphs — no more wall of text

The Word version uses real heading styles, so it drops straight into your own editing workflow. The PDF embeds a Unicode font, so smart quotes — and non-English text — render correctly.

Generated PDF title page with summary and tags
The title page: a brief summary and keyword tags, all generated locally.
Generated PDF transcript page split into chapters and paragraphs
The transcript, split into chapters with clean, sentence-aware paragraphs.

Optional: tidy up the grammar

Spoken language is messy. If you’d like the document text cleaned up — punctuation, capitalization, the occasional transcription slip — flip on Enhanced documents. CastPolish runs each paragraph through your local Ollama model with a strict guardrail: if the cleaned-up text drifts more than 15% in length, it throws the change away and keeps the original. Your words stay your words.

Still completely local

Like everything in CastPolish, this happens entirely on your own machine. The summaries and chapter titles come from a local LLM (Ollama), and the documents are built offline with the open-source fpdf2 and python-docx libraries. Nothing is uploaded anywhere.

See it in action

For this release I ran President Reagan’s 1986 address on the Challenger disaster — a public-domain recording, about four and a half minutes — through CastPolish. Start to finish, including audio cleanup, transcription, summaries, chapters, and both documents, it took 55 seconds.

Upgrading

PDF and Word export rely on two small libraries, fpdf2 and python-docx. If you run CastPolish from source, pull the latest and run pip install -r requirements.txt. If you use the app, open the Dependencies panel and click Install next to fpdf2 and python-docx — no restart needed.

CastPolish is free and open-source under the MIT license. Grab it on GitHub.

Albert Collver

Albert Collver

Albert Collver, Ph.D. — Lutheran theologian, podcaster, and lifelong coder since the '90s shareware era. I build free, open-source tools like CastPolish, liturgical software, and family-history projects. All local, no cloud.
Missouri