Is there a video editor that works offline with speech to text?

Yes. Video by Zaweba recognises speech on your own PC using whisper.cpp, an offline speech recognition engine, and uses it for two jobs: captions from the dialogue and a transcript you can edit the video with. Nothing is uploaded, there is no account, and it keeps working without an internet connection.

Most automatic captions in video tools are made by sending your sound to a server. That is quick, but it means your recordings leave your computer, it needs a connection, and it usually comes with a sign-in or a plan. Doing the recognition locally avoids all three.

Captions from speech

Video by Zaweba works through the dialogue voice line by voice line and turns it into captions. If you have a script, the script's words are lined up with what was heard, so names and unusual words are spelled as written rather than guessed. Captions are kept to at most two short lines, which suits phones as well as big screens.

Styles change how they look, and One word at a time splits captions into single words in capitals that pop in one by one, the style common on short videos. A caption can also be turned into a title clip if it needs special treatment.

When you are done, burn the captions into the picture or export them as a separate .srt file. More caption formats are supported too, including WebVTT, SCC, TTML, DFXP, EBU STL and MCC, and a timeline can hold several caption tracks, for example one per language.

Text-based editing

The Transcript tab turns speech into text in pieces of about 30 seconds. Then you can edit the video by editing the words: select words and press Delete and they are cut from the edit. Remove pauses tightens the gaps, Remove filler words takes out the ums and ers, find jumps to a phrase, and Save text writes the transcript out.

For interviews and talking-head videos this is often the fastest way to make a rough cut. Read the transcript, delete what you don't want, then fine-tune on the timeline.

Find speakers groups the words by who says them, again on your PC with no script. Click a speaker's heading to rename them everywhere.

Languages

English works out of the box. Speech can be recognised in 27 languages or detected automatically, but anything other than English needs a multilingual whisper.cpp model file added with Add model. Video by Zaweba doesn't download it for you, which is in keeping with the app not reaching out to the internet.

What to expect

Speed depends on your PC, since the work happens on it rather than on a server farm. Clear speech recorded close to the microphone gives the best results. Noise reduction and voice EQ can tidy up a recording before captions are made.

Accuracy is good, but read the captions through before publishing. Supplying a script, when you have one, is the best way to get names right first time.

Who it suits

Anyone whose recordings shouldn't leave their PC: interviews, meetings, schools, health and legal work, or simply people who prefer privacy. Creators who caption every video and don't want another subscription. And people working where the internet is slow or absent.

Video by Zaweba is free software for Windows under its EULA. It includes the speech model, which is one reason the download is large, and it works offline from the start.