Skip to main content
Send an image of the speaker in their setting when you create an instant voice, and the voice takes on the acoustics of that scene. A quiet studio keeps it clean; a beach or a factory floor adds matching ambience. The speaker’s accent and phrasing stay the same.

Listen

One recording, made in a meeting room:

Source recording

The same voice in four scenes, each reading this line:
Same voice, different scene. Listen to how it changes. Still sounds like me, right?

Studio

Beach

Factory floor

Car

Each still is a frame from that scene’s video.

Create a voice for a scene

Upload the recording and the scene image, then pass both to POST /v3/models/audio/voices:
Wait until the voice is ACTIVE, then pass its voice_id to Text to Speech. Every speech request with that voice uses the scene.

The image

  • A JPEG, PNG, or WebP that shows the speaker and their surroundings, as an asset_id, a public HTTPS url, or base64.
  • Send image on its own, without gender or age. Without an image, HeyGen picks a default scene from the speaker’s gender and age, detected from the recording or taken from those two fields.
  • The scene is part of the voice. For another scene, create another clone from the same recording with a different image.

Tips

  • To compare scenes, keep the recording, similarity, remove_background_noise, text, language, and expressiveness_boost the same, and change only the image.
  • Use a script that fits any setting, so the image is the only difference you hear.