Overcaption, a free subtitle studio for mac/windows built on whisper.cpp #3961

First of all thanks for whisper.cpp, my app would not exist without it.

I built a free desktop subtitle editor that uses whisper.cpp for the transcription part: word-level timestamps go through a segmentation step (reading-speed and language specific grammar rules) to produce subtitle cues, then there is a proper editor to fix text and timing on a waveform, plus burn-ins.

Some things that took me a while to get right, in case they are useful to others:

• Silero VAD pre-pass, then remapping token times back to the original timeline. I validate the remap against a silence map of the source audio. On files with music beds the energy-based map can come back empty and a stale delta compounds over the whole file.
• Hallucination filtering: -mc 0, dropping junk/annotation segments, gating on token probabilities, collapsing repetition loops. On a music/speech/ silence torture file this got us to zero hallucinated cues.

The app is Overcaption (https://overcaption.com), free and offline, no account. It's not open source (sorry), it's a product with some paid pro features, but the core is free with no time limit.

If there is a list of apps using whisper.cpp somewhere I would be happy to be added. And happy to answer anything about the VAD/timing part.