similarity sets how closely an instant voice follows the recording it was cloned from. Higher levels keep more of the speaker’s character, and more of the recording’s room sound. Lower levels give the voice more room to adapt its delivery. You set it once, when you create the clone with POST /v3/models/audio/voices.
Listen
One 45-second recording, cloned four times with a differentsimilarity each time. All four clones speak the same script with heygen-voice-1, in English, with expressiveness_boost: 0.
Source recording
Welcome back. Today, we will walk through your project settings, review the latest changes, and make sure everything is ready before you share your next video.
medium
high
extra_high
max
Create a clone at a level
ACTIVE, then pass its voice_id to Text to Speech.
An instant voice keeps the level it was created with. To hear another level, create a new clone from the same recording; each clone has its own voice_id.
Tips
- Record in a quiet room without echo.
extra_highandmaxkeep more of the recording’s room sound thanmedium. - To compare levels, keep the recording, text, language, and
remove_background_noisethe same, and send the sameexpressiveness_boostin every speech request. Try a few short passages. - For the closest match to the original speaker, use
maxand sendexpressiveness_boost: 0.0. - For HeyGen videos, our preferred combination is
maxwithexpressiveness_boost: 0.5. - An
imageof the speaker’s setting changes the voice’s ambience. Keep it the same, or leave it out, while you compare levels.

