I’m adding speech transcription to a small video-editing tool and need users to jump from a line of text to the matching moment in the clip. Has anyone used OpenAI’s transcription API for word-level or segment-level timestamps?

I’m curious how consistent the timing is with fast speech, pauses, and noisy recordings, and whether the available timestamp detail depends on the model or output format. Is it accurate enough for editing, or should I plan to align the transcript with the audio separately?