Transcribe video to text with AI

From video to a usable transcript

A raw speech-to-text dump is hard to read: no paragraphs, run-on sentences, “um”s. SubtitleWave builds the transcript from timed segments, so it can break paragraphs where the speaker actually paused (any gap longer than 1.5 seconds) and keep timestamps if you want them. Turn on “Timestamps in the transcript” in the Export tab for lines like [4:12] and that's why the budget moved, which is the format podcast show notes and interview research use.

The “Fix punctuation & misheard words” AI tool is worth one click on long recordings: it corrects capitalisation, adds missing full stops and fixes homophones (“their” vs “there”) without paraphrasing what was said.

Good uses for a video transcript

• Repurposing: turn a talk or YouTube video into a blog post, newsletter or social thread.
• Meetings and lectures: search a recording for the moment something was decided, then jump there with the timestamp.
• Interviews and research: quote people accurately; only the audio is sent, once, to Vercel AI Gateway for transcription, and it is not stored afterwards.
• Accessibility: WCAG 2.2 success criterion 1.2.1 asks for a text alternative for pre-recorded audio-only content; a transcript satisfies it.

Tips for a more accurate transcript

• Choose the spoken language instead of auto-detect when a video mixes in a few words of another language.
• Replace recurring misspellings of names in one go with Find and Replace.
• Speaker names are not detected automatically; add “Name:” at the start of lines where it matters.