Skip to main content
CLAIM26
French-language ElevenLabs expressive-voice cover featuring a studio microphone and colourful audio waveform.
ELEVENLABS / AUDIO TAGS

ElevenLabs Audio Tags: How to Control Emotion and Delivery in v4

Audio Tags are short instructions in square brackets that guide how ElevenLabs delivers a line. Use them to shape emotion, vocal delivery and reactions in Eleven v4, v4 Turbo and v3, then listen to a short sample before producing a longer recording.

Artificial IntelligencePublished 9 min read
Create expressive AI voiceovers with ElevenLabsAffiliate disclosure: Claim26 may earn a commission from some ElevenLabs links at no extra cost to you. Pricing and product claims are checked against official sources.
GOALA short, natural script with clear delivery directions, tested on the voice and model intended for the final recording.Pricing and product details checked against official sources on November 22, 2026.
French-language ElevenLabs expressive-voice cover featuring a studio microphone and colourful audio waveform.
KEY TAKEAWAYAudio Tags are short instructions in square brackets that help ElevenLabs shape how a voice speaks. Learn which models support them, how to place and combine tags, and how to avoid unnatural results.

What are ElevenLabs Audio Tags?

ElevenLabs Audio Tags are short, natural-language instructions placed inside square brackets in a text-to-speech script. They guide the way the model delivers the words that follow, for example by asking for a whisper, a burst of excitement, a laugh or a more restrained vocal tone.

A tag is a performance direction, not part of the sentence you want the listener to hear. For example, [whispers] I think someone is outside. asks for a quieter delivery, while [sighs] I thought we were finished. adds a vocal reaction. The exact result still depends on the model, the selected voice and the surrounding context.

Which ElevenLabs models support Audio Tags?

Eleven v4 and Eleven v3 both support Audio Tags, and ElevenLabs recommends v4 for most new text-to-speech work. The same basic tag syntax works with both models, so an existing v3 script can be tested with v4 without rewriting every instruction.

Eleven v4 Turbo also supports Audio Tags and is intended for real-time applications where response speed matters. For a recorded voice-over, audiobook or carefully edited narration, start with Eleven v4. For a live or interactive voice experience, compare v4 Turbo against the response-time requirements of your application.

The main types of Audio Tags

Audio Tags can describe several parts of a performance: emotion, vocal delivery, human reactions, pacing, accents, character qualities and even non-verbal sound effects. The categories overlap, so treat them as useful ways to think about your script rather than as a rigid technical menu.

The official examples include [curious], [crying] and [mischievously] for emotional direction; [whispers] and [shouts] for delivery; and [laughs], [clears throat] and [sighs] for reactions. ElevenLabs also supports descriptive instructions written in ordinary language, so the examples are not a closed list.

Emotion tags: give the line an intention

An emotion tag helps the model understand the attitude behind a sentence. Use [curious] when a character is genuinely asking a question, [excited] for an enthusiastic announcement, [crying] for a distressed line, or [mischievously] when the speaker should sound playfully deceptive.

The words around the tag still matter. A cheerful sentence followed by [crying] may produce an inconsistent performance because the emotional instruction conflicts with the meaning. Write the dialogue first, then add the emotion that a listener should understand at that moment. The tag list is not exhaustive, but descriptive or experimental instructions can behave differently across voices.

Delivery tags: control how the voice speaks

Delivery tags focus on the sound and intensity of the voice rather than the character's emotional state. Examples include [whispers], [shouts], [softly], [quietly] and [low, gravelly voice]. They are useful for creating contrast between a normal line, a secret, a warning or a dramatic announcement.

You can combine delivery with emotion when the two instructions add useful detail. For example, [whispers, worried] Don't turn around. describes both a quiet voice and an anxious state. Avoid stacking several near-identical directions, because a long list of instructions can make the intended performance less clear.

Human reactions and sound effects

Reaction tags add vocal events that are not ordinary spoken words. Examples include [laughs], [sighs], [gasps], [clears throat], [exhales] and [snorts]. They can make dialogue feel less like a continuous reading and more like a person reacting to what is happening.

Eleven v4 also supports environmental sound-effect tags such as [gunshot], [applause], [clapping] and [explosion]. These request an audible event rather than a change in vocal delivery, so use them deliberately and test the result. The Enhance button is designed to add vocal directions, not music or unrelated sound effects.

How to add Audio Tags in the Text to Speech interface

Open Text to Speech in ElevenLabs, choose a voice and select a compatible model such as Eleven v4. Write a short passage in the text editor before adding tags. A few natural sentences make it easier to judge whether the voice and accent suit the project.

Insert each instruction in square brackets immediately before the passage where the delivery should change. Generate a short sample, listen to the complete transition and adjust one instruction at a time. The names and arrangement of controls in the interface can change, so follow the current model selector and editor shown in your account.

Use Enhance to suggest Audio Tags automatically

The ElevenLabs Text to Speech interface includes an Enhance button that can automatically add relevant Audio Tags to your input text. The feature uses an AI language model to suggest vocal directions while preserving the original dialogue and its meaning. Treat the result as a first draft, not as an instruction set that must be accepted unchanged.

Review the suggested tags before generating audio. Remove directions that contradict the line, describe something that cannot be heard, or add a sound that you do not want. ElevenLabs' built-in guidance focuses on vocal tags and warns against instructions such as [standing] or [grinning], as well as tags for music or unrelated sound effects. Generate a short sample after editing.

Combining tags, context and punctuation

You can combine multiple directions in one pair of brackets, separating them with commas. For example, [whispering, playful] This is our little secret. asks for a quiet delivery with a playful tone. A short combination is usually easier to interpret than a long list of adjectives.

Punctuation is another important part of the prompt. Ellipses can suggest a pause or hesitation, capital letters can add emphasis, and sentence structure helps the model understand the rhythm. Place a new tag when the intended delivery changes, rather than repeating the same instruction mechanically before every sentence.

Audio Tags in dialogue with multiple speakers

Audio Tags can be used in dialogue, but a tag does not automatically create a new speaker. For a conversation between two or more characters, assign a voice to each speaker using ElevenLabs' dialogue-generation workflow, then add delivery instructions inside the text for that speaker.

For example, one speaker might say [excited] We finally got the message! and the other might answer [relieved] I was beginning to worry. The voice ID identifies who is speaking; the tag guides how that speaker delivers the line. Listen to the transitions and check that each character remains recognisable.

Using Audio Tags through the ElevenLabs API

Audio Tags are not limited to the web editor. They can be included in text sent to the ElevenLabs Text to Speech API. The documented Create Speech and Stream Speech endpoints accept text and a compatible model ID, including eleven_v4 and eleven_v3. The v4 Turbo model ID is eleven_v4_turbo.

For multi-speaker conversations, ElevenLabs also documents Create Dialogue and Stream Dialogue endpoints. Keep the same square-bracket syntax in the text you send, choose the correct voice for each speaker and verify the returned audio. Always check the current API reference for request fields, model availability and limits before integrating the feature into an application.

Audio Tags versus SSML break tags

SSML is a markup language used by some speech-synthesis systems to control elements such as pauses and pronunciation. However, Eleven v3 and Eleven v4 do not support SSML break tags. A snippet such as <break time="1.5s" /> is therefore not the right way to insert a pause with these models.

For v3 and v4, use punctuation, text structure and suitable Audio Tags to influence pacing and delivery. Other ElevenLabs models have different capabilities, so do not assume that a pause instruction documented for Multilingual v2 or another model will work with v4.

Why an Audio Tag may not produce the result you expect

A tag guides the model; it does not guarantee an identical performance on every generation. A delivery that is already part of a voice’s natural range is generally easier to reproduce. Eleven v4 follows tags more reliably than earlier models, but a voice that has never whispered or shouted may still produce a less convincing result. Test the wording with the exact voice and use case.

When a result sounds wrong, keep the voice and sentence unchanged and test a simpler tag. Then try a more explicit description of the sound you want, such as [low, gravelly voice]. Avoid changing the model, voice, script and several settings at the same time, because you will not know which change affected the result.

Three Audio Tag scripts you can adapt

Suspense narration: [low, steady voice] The corridor was completely silent. [whispers] Then I heard someone call my name. [sighs] I knew I wasn't alone. This example uses a restrained opening, a quieter reveal and a final reaction; adapt the wording to the voice you have selected.

Promotional voice-over: [confident] Your next idea deserves to be heard. [excited] Create, publish and share your message with a voice that fits your brand. [warmly] Start with a short sample and build from there. These are sample instructions, not a recording or test performed by Claim26.

Two-person dialogue: Speaker 1: [curious] Did you hear that? Speaker 2: [whispers] I think it came from the next room. Speaker 1: [nervously] Let's take a look. Assign a separate voice to each speaker in the dialogue workflow; the labels in this example simply explain the structure.

Voice cloning, credits and responsible use

Professional Voice Clones are supported in Eleven v4; ElevenLabs notes that they are not fully optimised for Eleven v3 and recommends trying v4 for this use. Audio Tags do not grant permission to imitate someone. Use a voice you own or are authorised to use, follow ElevenLabs' current terms and avoid presenting generated speech as an authentic recording of a real person when that would mislead listeners.

Generation costs and commercial-use rights depend on the current plan and applicable terms. Before producing a long narration, check the model, the credit estimate and the licence that applies to your intended use. A successful export by itself does not confirm that you have every right needed to publish the audio.

A final checklist before exporting

Listen to the complete recording, not only the first tagged sentence. Check that the voice pronounces names correctly, that emotional changes make sense, that reactions are not excessive and that the volume remains comfortable. For dialogue, make sure each speaker stays consistent and that the turn-taking is clear.

If a passage sounds artificial, remove unnecessary tags before adding more. A small number of well-placed directions usually gives you more control than annotating every word. Save the version that works and confirm credits, usage rights and the final audio format before publishing.

Conclusion: use fewer tags, placed more deliberately

ElevenLabs Audio Tags make it easier to direct an AI voice beyond a neutral reading. With short instructions in square brackets, you can guide emotion, delivery, reactions and sound effects in Eleven v4, v4 Turbo and v3. The model still relies on the selected voice, the surrounding text and the clarity of the direction.

For a new project, begin with Eleven v4, write a natural script and test a short passage. Add one useful tag where the performance needs to change, listen to the result and refine only what is necessary. That simple process keeps the narration expressive without making it sound over-directed.

BEFORE YOU CHOOSE

Check these four signals.

✓Eleven v4 selected for most expressive narration, or v4 Turbo for real-time use

✓Tags placed immediately before the passage whose delivery should change

✓Voice, accent and context checked with a short sample

✓Conflicting or repetitive instructions removed

✓Credits, voice permissions and commercial-use terms checked before export

MOVE FROM IDEA TO TESTING

The safest way to choose is to build a small real-world workflow.

Test the part of the customer journey that matters most before moving your entire system.

Create expressive AI voiceovers with ElevenLabs
PRIMARY SOURCES

Official sources for this guide

Software pricing and features can change. Review the official documentation below before choosing a product or plan.

READ NEXT

Continue with the question that is most likely to affect your next decision.

Cinematic audio-production console with four copper-lit controls, microphone, official white ElevenLabs logo and prominent Master Your Voice title.
Artificial Intelligence · 13 min readElevenLabs Voice Settings Explained: Stability, Similarity, Style and SpeedRead the guide →
ElevenLabs Sound Effects illustrated in an audio production studio, with prompt-based effects, waveforms and sound categories.
Artificial Intelligence · 16 min readElevenLabs Sound Effects: Generate AI Sound Effects for Video and GamesRead the guide →
Cinematic voice-generation studio with microphone, audio waveform, developer code icon, single VOICE VIA API headline and the genuine full ElevenLabs wordmark on light background.
Artificial Intelligence · 14 min readElevenLabs Text-to-Speech API: Generate Your First AI Voice in Python & JavaScriptRead the guide →
MOVE FROM READING TO TESTING

Build your first page and test the offer in a real sales journey.

Systeme.io lets you start free and move up only when the business genuinely needs more capacity.

systeme.io · Start free