Speech-to-text is the job: you want a transcript. ASR is the recogniser that does the first pass. Forced-alignment is a later, easier pass that times a known transcript rather than guessing the words. Caption pipelines run ASR and alignment together, which is why a wrong product name and a late product name feel like one bug. They are not. One is recognition. One is the clock.
Errors cluster where the audio is hard: overlap, music, jargon, proper nouns, a second language. Isolating the vocal before you recognise is the cheap accuracy win. Proofread names before you burn; the one word ASR misses is usually the one on the label. Closed captions and burned-in captions both inherit that transcript. Karaoke then needs word timestamps on top, which some ASR stacks emit and some do not — if they do not, align the corrected text.
Platform-native captions are someone else's ASR. You cannot correct TikTok's recogniser in your editor. If the mute feed has to show the right spelling, that is a burn of a transcript you actually read, not a hope that the app's engine knows your SKU.
In practice
- Proofread names, numbers and product spellings before you caption. ASR will not know your SKU.
- Separate recognition mistakes from timing mistakes before you offset a whole track.
- Do not treat in-app auto captions as a checked ASR pass. You do not own that engine.
The mistake to avoid
Publishing the raw recogniser output as captions. The transcript is a draft. The burn is the publication.
Where you will run into it
- Get a Video Transcript — Turn spoken words into a written script.
- Add Subtitles Automatically — You upload the video. The words appear, timed to the voice.
- AI Caption Generator — Drop a video, get timed captions. Free, on your device, no upload.
Related terms
Speech-to-text
Speech-to-text meaning: converting spoken audio into written text, the transcript captions, translation, and search all depend on.
Caption timing
Caption timing is when each line appears and disappears relative to the speech — the clock, not the words and not the look.
Closed captions
Closed captions are a separate subtitle track the player draws, which the viewer can switch off and which each platform styles itself.
Text-to-speech
Text-to-speech converts written text into spoken audio using a synthetic voice you choose before generating.
Voice cloning
Voice cloning meaning: building a reusable synthetic voice from a real sample so new scripts can be spoken in that voice later.
The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync — in your browser or on your phone.
Free on iPhone. On a computer? The same account works on Versely Web, no install needed.