An audio tag is an English word or phrase in square brackets in text. It changes how the voice speaks, or adds a sound such as a laugh. Public and instant voices perform audio tags on all three models: heygen-voice-1, heygen-voice-1-turbo, and heygen-voice-1-turbo-free. Both speech endpoints accept them. Professional voices do not perform audio tags.
Each tag is in one of four categories: emotion, delivery, acoustic event, or scenario. The common tags table shows what each category does.
Add tags automatically
To add tags to a script you already have, give this prompt to an LLM, with your script in place of<your text>. It uses only the common tags.
Prompt
Listen
Every sample below is the exacttext shown under its player. Unless a section names a different voice, one instant voice with the same settings speaks every sample.
Same line, different tags
One script, first with no tags and then with a different lead tag. The words stay almost the same, and the tags change the delivery.No tags
[calm]
[excited]
[nervous]
[warm]
[whispers]
One category at a time
Emotion, delivery, acoustic event, and scenario tags, each in its own script.Emotion: nervous, relieved, excited
Delivery: whisper, loud, fast
Delivery: pause and hesitation
Acoustic events: laugh, sigh, gasp
Acoustic events: gasp and laugh
Scenario: news anchor to sports commentator
Scenario: sharing a secret
Example scripts
These scripts mix categories, with one tag at each position.Telling a story to a friend
Customer support
Ad read
Tutorial
Bedtime story
Accent preservation
A cloned voice keeps its accent when you add tags. The voice below was cloned from a speaker with an Indian English accent. The first player is a neutral reading with[matter-of-fact]. The other players are tagged lines spoken by the same voice.
[matter-of-fact]
[shy]
[talking to a small child]
[surprised]
Emotion intensity
To make an emotion weaker or stronger, putslightly or extremely in front of the emotion tag, such as [slightly sad] or [extremely angry]. This works with the emotion tags in the Feelings and Tone lists of All supported tags. Each group below is one line spoken by one voice, first with no tag and then at three levels of the same emotion. The two tabs are two voices with the same lines.
- Voice 1
- Voice 2
SadHappyAngry
No tag
[slightly sad]
[sad]
[extremely sad]
No tag
[slightly happy]
[happy]
[extremely happy]
No tag
[slightly angry]
[angry]
[extremely angry]
How tags work
- An emotion, delivery, or scenario tag changes the speech after it. It has an effect for up to approximately 17 seconds of speech, approximately 233 characters of English. To keep the effect for a longer time, write the tag again in a later sentence.
- An acoustic event tag, such as
[laughs]or[sighs], adds one sound at its position. - These delivery tags are exceptions. Each one acts one time at its position, as an acoustic event does:
[pause],[hesitates],[after a long pause],[voice cracking],[trailing off], and[catching themselves mid-sentence]. - When
textcontains a tag other than[pause],expressiveness_boostis not applied. The request is not rejected.
Writing tags
- Write each tag in lowercase, exactly as it is in the list below, with no closing tag.
[deadpan, dry humor]is one tag.[sad, quietly]is not a tag. - To make an emotion weaker or stronger, write
slightlyorextremelyand one space in front of an emotion tag from the Feelings or Tone list in All supported tags, such as[slightly sad]or[extremely happy]. Other words, such as[very sad], and other tags, such as[slightly whispers], are not tags. - Put one tag at a position. Do not put two tags side by side, such as
[angry] [whispers]. - Write tags in English, also when
textis in a different language. - Bracketed text that is not a tag, such as
[Section 4], is usually not spoken. To have those words spoken, remove the square brackets. - Tags count toward the 10,000-character limit of
text. - Markdown is not interpreted.
**word**does not add emphasis, so remove Markdown fromtextbefore you send it.
Common tags
Choosing a voice
For an instant voice, clone from clear speech recorded in a quiet room without echo. See Instant Voice Clone. The same tag can sound different from voice to voice. Try a tag with a few voices before you choose one.Tips
- A request without
seeduses a random one, so sending the same request again gives a new take. Send the sameseedto try to repeat a take; repeats are best effort. To steer a different performance, change the wording or punctuation near the tag and send the request again. - You can mix tags from different categories in one text.
- Choose tags that fit the words.
- Put an acoustic event where the sound happens, such as
[laughs]after the funny line. - To change the mood inside a sentence, use a mid-line change tag, such as
[rising panic],[suddenly serious], or[voice dropping to a whisper].

