Word-level timestamps for transcription

I want word-by-word timestamps (word-level granularity) for the transcription API, not just phrase-level chunks. This is important for creating precise word-by-word subtitles for videos.