The voice is in the file. The mouth is not. That is allowed, and it is common: a product plate with a read, a joke whose punchline is a line from off camera, a second person answering from outside the shot. Voiceover is the editorial choice to lay a read under picture on purpose. An off-screen speaker is the diegetic or documentary version — someone in the scene you cannot see. The caption job is the same either way: a mute viewer has words with no face to attach them to.
Speaker ID is the prefix that names who is talking — [Maya], Host: — in the same cue as the line. SDH assumes you cannot hear the mix, so the ID is required when the speaker is new, overlapping, or off screen. Auto captions transcribe speech and will not reliably invent those labels. Karaoke still needs the ID on the first token of a new speaker, or the accent walks across a line whose owner is unknown.
Do not generate a photoreal presenter to 'give the voice a face' unless the brief asked for a talking head. Health, legal, and finance spots that want a read without a synthetic performer use the split on purpose: picture of the product, voice off screen, captions with an ID. Lipsync is the other door, and it is only the door if a mouth has to match.
In practice
- Add a speaker ID when the mouth is not in frame. The words alone do not name the source.
- If the brief is a product plate plus a read, generate the read as voiceover or TTS. Do not invent a host.
- Split overlapping off-screen lines into two cues rather than stacking two names on one block.
The mistake to avoid
Generating a talking head because the script has a voice. Off-screen speech is a mix and caption problem, not a casting problem.
Where you will run into it
- Add Captions to a Video — Speech in, styled subtitles out — no manual timing.
- Add a Voiceover to a Video — Type the script. Get a narrated video back.
- Add Multi-Speaker Dialogue to a Video — A whole cast, one API call.
- AI Caption Generator — Drop a video, get timed captions. Free, on your device, no upload.
- AI Text to Speech — One script, several engines, one bill in credits.
Related terms
Speaker ID
A speaker ID is the caption label that names who is talking — [Maya], Host: — so a mute viewer can tell overlapping or off-screen voices apart.
SDH
SDH is a caption track that includes speech and non-speech audio — [door slams], [music], speaker IDs — so the file is usable without sound.
Voiceover
A voiceover is a spoken track laid under picture, rather than lipsync that drives a mouth or a caption track that visualises speech.
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 the web and on iPhone, one account.