Long audio stays on its original timeline
When audio transcription is needed, the service processes the audio in windows of up to 30 seconds with a 2-second overlap. It adds each window’s actual start time back to its transcript segments and removes repeated wording at the joins. The timestamps therefore refer to positions in the full video, rather than restarting at zero for every window.
Caption retrieval and audio transcription are different paths
The service reads an available TikTok caption track first. Audio transcription is used only when the public video metadata confirms that no native caption track is present. If TikTok lists captions but they cannot be read, the request reports that failure instead of silently substituting a generated transcript.
Audio recognition is generated by Cloudflare Workers AI and may contain errors. A failed later audio window does not count as a complete transcript.
Use the timed result
Open the tool, paste a public video link, and choose Timestamps to inspect each segment’s start and end time. Download SRT for a numbered subtitle file or WebVTT for a WEBVTT subtitle file. Both retain segment timing from the source captions or the audio transcription.