Cartesia Audio Tags: 58 Emotions and XML Syntax Guide
Cartesia Sonic 3.6 uses XML-style tags to request emotions and speech controls. An emotion cue such as <emotion value="excited"/> goes before the words it should affect. The value must be a documented emotion name, not a new label you invent.
This guide is based on upstream Cartesia documentation. The examples have not been runtime-tested in Zoice, and the illustrated composer controls do not prove that every upstream tag passes through the app integration. Cartesia describes emotion control as English-only and beta, with Emotive voices recommended. That limitation applies to the emotion feature, not a claim that the whole model can only speak English.
Start With One Emotion Tag
Place the emotion cue immediately before a short sentence whose meaning fits the requested delivery. Start with one emotion rather than mixing several conflicting cues. For example:
<emotion value="excited"/> We are going on a trip tomorrow!
The tag is an instruction, not text you intend the voice to speak. Preserve the angle brackets, quotation marks and self-closing slash. If you use Cartesia Sonic 3.6 in Zoice AI Voice, select that model before entering Cartesia syntax. A tag written for the wrong model is not a meaningful compatibility test.
All 58 Documented Cartesia Emotion Values
The documented values are listed below in full. Use the exact value spelling inside the emotion attribute:
neutral, calm, angry, content, sad, scared, happy, excited, enthusiastic, elated, euphoric, triumphant, amazed, surprised, flirtatious, curious, peaceful, serene, grateful, affectionate, trust, sympathetic, anticipation, mysterious, mad, outraged, frustrated, agitated, threatened, disgusted, contempt, envious, sarcastic, ironic, dejected, melancholic, disappointed, hurt, guilty, bored, tired, rejected, nostalgic, wistful, apologetic, hesitant, insecure, confused, resigned, anxious, panicked, alarmed, proud, confident, distant, skeptical, contemplative, determined.
Emotion names are not interchangeable descriptions of intensity. For instance, calm and excited request different directions; adding both does not create a reliable intensity slider. Choose a value that matches the sentence and evaluate any output with the selected voice. The documented list does not imply that all voices express every emotion equally.
Syntax Reference: Emotions and Other Controls
| Control | Copyable example | What to check |
|---|---|---|
| Emotion | <emotion value="calm"/> Let us take this one step at a time. | Use a documented value before text; English-only beta emotion control. |
| Laughter | That was not the plan. [laughter] | Laughter uses square brackets, unlike the XML-style controls. |
| Speed | <speed ratio="1.2"/> Here is the short version. | Use a single numeric ratio, not prose or a range in the attribute. |
| Volume | <volume ratio="0.5"/> Let us keep our voices down. | A numeric volume ratio is not an emotion name or a request for a whisper. |
| Break | First, check the address. <break time="1s"/> Then continue. | The time attribute includes a duration unit. |
| Spell | Your reference is <spell>ABC123</spell>. | Wrap the characters with opening and closing spell tags. |
These controls are documented separately. A slower sentence is not automatically sad, and a softer sentence is not automatically calm. Begin with the control that corresponds to the change you actually want. Ordinary fillers such as uh and um can be written as script text; they are not custom XML commands.
Copyable Cartesia Script Examples
An emotion change in a short story
<emotion value="excited"/> We are going on a trip tomorrow! <emotion value="calm"/> First, let us pack slowly. <emotion value="sad"/> I will miss the garden. <emotion value="happy"/> But I cannot wait to see the sea. [laughter]
Each emotion precedes the corresponding sentence. The example demonstrates intended placement, not an observed performance. Do not assume a cue changes only one word or remains active for an exact duration unless the integration documentation specifies that behavior.
A calmer explanation with a break
<emotion value="calm"/> Let us review the instructions together. <break time="1s"/> Start with the first item, then move to the next.
Speed and volume as separate trials
<speed ratio="1.2"/> The doors open at nine. Please have your ticket ready.
<volume ratio="0.5"/> The library is closing soon. Please finish your work quietly.
These are two separate example scripts. Keeping the controls separate makes it easier to identify whether a change is attributable to speed, volume, wording or voice selection. No output from these examples was generated for this guide.
Entering Cartesia Syntax in Zoice AI Voice
Open app.zoice.com/voice and choose Cartesia Sonic 3.6 when available in the model selector. Confirm the selected model before entering any XML-style text. The screenshots below are existing close-ups of shared composer controls; they do not show Cartesia selected or demonstrate successful Cartesia output.
1. Enter the script in the composer
This crop shows the Enter the script here field. Use it to enter a short Cartesia example after selecting the matching model. Keep the literal XML-style punctuation in the script; this guide has not tested how the app processes it.
2. Choose a compatible voice
The existing crop shows Select voice, not a Cartesia-specific voice list. Use the voice options exposed for your selected model. Cartesia recommends Emotive voices for emotion control, but this guide does not claim the Zoice picker exposes an Emotive filter or that every listed voice supports every control.
3. Review before requesting audio
Check the model, value spelling and script punctuation before using Generate. If you choose to generate audio, listen for the requested change rather than assuming success from the presence of a tag. This article does not show a generated player, a download workflow or a tested result.
Common Syntax and Expectation Mistakes
An invented emotion label is not part of the documented value list. Choose the nearest documented emotion instead of assuming Cartesia accepts free-form emotion names. Avoid curly quotation marks copied from a word processor. Keep ratio values numeric: <speed ratio="1.2"/> and <volume ratio="0.5"/> are examples of valid attribute shapes; a written range inside either attribute is not.
If a control appears to be ignored, separate syntax problems from integration questions. Recheck the chosen model, try a single control with consistent wording, and consult current Zoice developer documentation for exposed options. Without an actual runtime test, this guide cannot identify whether a mismatch comes from the model, voice or app.
FAQ
Can I create my own Cartesia emotion value?
Arbitrary custom emotion values are not a documented feature. Use one of the 58 listed values. You can rewrite the surrounding sentence, but that does not make an invented XML value supported.
Are the emotion tags multilingual?
Cartesia documents emotion control as English-only and beta. Do not infer multilingual emotion control from the model's broader language capabilities.
Is laughter an XML tag?
The documented laughter cue is [laughter], using square brackets. Emotion, speed, volume and break use XML-style syntax, while spell wraps text with an opening and closing tag.
Does a documented tag guarantee it works in Zoice?
No. Upstream documentation describes model controls, not a completed app-integration test. This guide has not generated Cartesia audio in Zoice or verified pass-through for these controls.
Open the Composer or Developer Documentation
Start in Zoice AI Voice to inspect the available model and voice options. For API integration, consult Zoice developer documentation rather than assuming all upstream controls are exposed by every endpoint.