AI clean-up for transcripts

AI clean-up is an optional last step for file transcripts. It uses a large language model from a provider you choose (Google, OpenAI or Anthropic), with your own API key, to polish the text a transcriber cannot: fix misheard words, sort out punctuation, rewrite fillers into prose, or summarize.

It applies only to file transcripts — never to dictation, Translate, Nikud or Speak, which stay entirely on your computer. It does nothing at all until you add a key.

Settings → AI cleanup. Pick a provider from the dropdown and paste a key:

• Google — Gemini Flash — recommended; a free key from aistudio.google.com is enough for most people.
• OpenAI — a key from platform.openai.com.
• Anthropic — Claude — a key from console.anthropic.com.

You can leave the model field blank to use the provider’s default, or type a specific model name. Keys are stored encrypted on this PC only, tied to your Windows account, and each has a Remove button.

• Clean up — fixes misheard words, spelling and punctuation, keeping every sentence exactly where the speaker said it.
• Rewrite — turns the transcript into polished written text — same content, no fillers or stutters.
• Summarize — a faithful summary of what was said, in the same language.

The result appears in its own panel with Copy and Save. Long recordings are processed in sections so nothing is cut off. Yiddish output is written in the traditional spelling most readers expect.

On the Transcribe file page, switch the “Clean up with AI” pill on before starting and the clean-up runs as the last phase of the job — the timeline shows its progress like the others. Speaker names are kept, so an interview comes back tidy and still labeled.