Make subtitles from audio with AI
Transcribe local speech into timed segments for SRT or WebVTT, or format a supplied list of segments without transcribing the recording again.
Editors can make subtitles from audio with AI to produce a first caption draft from a lesson, interview or narrated demonstration. The speech-captions skill uses a local speech model and media tools to preserve timings alongside the recognised words.
A speaker in a recording might mention a product name that resembles an everyday word. With a timed draft, a reviewer can jump to that moment and correct the text. A separate route accepts plain prose, but timings estimated from text length need to be checked against the recording.
What it does
- Transcribe speech from an authorised local audio or video file.
- Detect the principal spoken language.
- Save captions as SRT for an editing workflow.
- Produce WebVTT for a compatible web player.
- Convert existing timed segments or estimate cue timing from plain text.
Install and use
- Install the skill, Python, uv and FFmpeg. The local faster-whisper model may need to download its weights before first use.
- Ask “Create an SRT draft for this lesson. Keep the times so I can check every product name.”
- Play the recording with the captions and review timing, names, terminology and any overlapping speech before release.
FAQ
Source
Made by timelapsetech. Original skill: speech-captions, published on GitHub under the MIT licence.
View on GitHub