Private transcription for Windows
Turn a podcast, video, or audio file into a clean, searchable transcript. Podcast Reader transcribes on your computer, keeps a local library, and gives long-form audio a page made for reading.
Windows installer in final prep. Signed public downloads are not available yet. Developers can build the app from source today.
The privacy boundary
Speech-to-text runs on your computer. There is no Podcast Reader account, and the project is open source.
Network boundaries are explicit: URL inputs are fetched from their source, transcription models download during setup, and transcript text is sent to a chapter provider only if you configure one. Local-file transcription itself does not upload your audio to a transcription service.
How it works
Paste a supported URL, choose a local audio file, or drop a file into the app.
Choose a speech model that fits your hardware. CPU works; an NVIDIA GPU is optional for faster transcription.
Open the finished transcript beside its audio, jump by timestamp, search for exact words, and copy what you need.
Made for reading
Podcast Reader turns timestamped speech into calm, readable pages. When you configure an optional chapter provider, it can add chapter summaries, key points, and pull quotes. Without one, the full transcript and timestamp navigation still work.
Search
Search highlights the matched term inside the transcript and keeps passage context around it. Unicode-aware matching works across scripts and languages, and next/previous controls make long conversations quick to scan.
Export
Copy the whole transcript or the current chapter as plain text or Markdown, with timestamps when useful. Export is generated in the page; it does not call a network service.
Speakers
Install the optional speaker-diarization pack to label speaker changes. The pack runs locally after installation and is never enabled by default.
Private web access
Enable Private web access to open the read-only library from a phone or laptop already on your Tailscale network. Podcast Reader uses Tailscale Serve—not Funnel—and keeps the engine bound to the desktop's loopback interface. Tailscale is optional and is not bundled.
Requirements
Start on a CPU. Add a GPU only if you want more speed.
Model and component downloads happen at runtime and show their size before installation. The Windows build bundles dynamically used, LGPL-configured FFmpeg/FFprobe binaries; license notices and component provenance ship with the app.
Get started
The source is public and the Windows build path is documented now. A signed public installer will appear here only after the release-signing work is complete—no unsigned installer is presented as a release.