Two ways to get an SRT file
If the audio already exists, such as a recorded interview, lecture, podcast, or the soundtrack of a video, the subtitle file has to come from transcription: software listens to the recording and writes down what was said, with a start and end time for each line.
If you are producing the voice yourself from a script, transcription is unnecessary. The words are already known, so the subtitle file can be built directly from the generated narration. That is faster and avoids transcription mistakes.
From a recording you already have
NarrateHQ's Speech to Text takes a finished audio or video file of up to 120 minutes and returns a timestamped transcript. It is batch, not realtime: you upload the whole recording and get the whole transcript back. The Free plan includes 10 minutes of transcription a month to try it, and Starter includes 30 minutes.
Choose the recording's language instead of relying on auto-detect, which can confuse similar-sounding languages. Before exporting, read the transcript for names, numbers, and technical terms, because those are where transcription errors cluster.
- Export as SRT for most video editors, VTT for web players, or plain text for notes.
- To re-voice an old recording, send the transcript into the studio as a script, edit it, and generate new narration.
From AI narration you generate
When NarrateHQ generates narration, it creates the SRT from the generated section spans and the final merged take. When a section is replaced, the subtitle file is rebuilt as part of the same publish path, so the downloaded audio and SRT always describe the same version.
This keeps the workflow useful for YouTube, course, podcast, and documentary edits where a subtitle file is part of the delivery package rather than an afterthought.
- Download the audio and SRT from the same completed generation.
- After a regeneration, download the refreshed SRT again.
Know what the timing represents
For generated narration, cue timing is derived from the audio spans available to the application. A multi-sentence section may use proportional timing inside that span, which is practical for editing but is not word-level forced alignment. For a transcribed recording, timing comes from the segments the transcription returns.
Either way, spot-check long cues in the target editor before publishing, and confirm that the final cue does not run past the end of the audio.
Keep the SRT in sync after edits
Most subtitle drift comes from editing the audio after the SRT was exported. Treat the audio and its SRT as one pair: whenever one changes, export both again and label them with the same revision.
- If a sentence in the script changes, regenerate the section that contains it, then download the new audio and SRT together.
- If you trim or move audio in the video editor, shift or re-time the subtitles there as well.
- Import the SRT where it will be published: CapCut accepts SRT files on desktop and web, and YouTube Studio accepts subtitle files on upload.