I’m working on subtitles and need to line up spoken words with the audio, not just get a block of transcript text. Does the OpenAI audio API return word-level timestamps, or is timing available only at a larger segment level? I’d also be interested in whether the response format changes which timing details are included.