Give ElevenLabs v4 an Expressive Face with Anam

ElevenLabs' new Eleven v4 voice model is their most expressive voice model yet. The models emotion tags support a wide range from being restrained, affectionate, or full of frustration. Once you pair that voice with Anam's Cara-4 expressive avatar model and Director Notes, you get a very expressive real-time interactive avatar.

Eleven v4 comes with two impressive features: greater emotional range, and the ability to keep a recognizable voice across languages while adopting a native accent.

More expressivity in every line

Consider the line, "I thought you'd come back."

Spoken softly, it can sound relieved. With a clipped delivery, it can sound accusatory. The words haven't changed, but the listener's understanding of the scene has.

Eleven v4 TTS improves how its model interprets tone, pacing, and context, as well as how accurately it follows natural-language direction and inline audio tags. That gives builders more control over the delivery of individual lines, including emotional shifts within a response.

In the following demo, we show how an interactive Anam avatar with the persona of an actor can recite a section of a Shakespearean play with deep emotion and expressivity.

You can use this voice model beyond the use-case of actor avatars. A training character, for instance, might become frustrated when a learner avoids a difficult question. A tutor avatar might use a gentler delivery after a mistake. These are moments where the manner of speaking carries part of the message.

Direct the Performance

Anam's Cara-4 model generates expressive avatar video with natural facial movement and lip sync. Director Notes give you a way to steer that performance as the conversation changes.

There are two levels of direction. The default is a session style that establishes the character's usual presence, such as warm or supportive. The second uses director cues, which introduce a temporary change during a spoken turn, such as curiosity, sadness, or anger. Together, they let you keep a consistent character while varying the response to a particular moment.

For the actor example, you might choose a restrained intro and add stronger direction at an emotional turn in the passage. The voice can build in intensity while the avatar receives a corresponding cue.

The Director Notes guide explains the available styles and cues. Its expressivity setting controls the strength of the avatar's performance direction.

Director notes guide broad performance rather than exact gestures. A cue asks for an emotional direction; it doesn't prescribe a particular eyebrow movement on a particular word.

One voice, many languages and accents

ElevenLabs says both Eleven v4 and v4 Turbo support more than 90 languages. A voice recorded in one language can speak another with a native accent while retaining its identity, with stronger accent adherence across the generation. All of this can take place during the same session, and does not require changing the voice IDs.

A recurring character benefits from that consistency. A language tutor, for example, can retain a recognizable vocal identity as the lesson moves between languages - switching between accents depending on which language it speaks. Similarly, a product guide can keep the same face and voice across localized experiences based on the user’s preferred language and accent.

For example, in the same demo of an actor powered by Eleven v4 and Anam, we can switch between having it recite a theatrical passage in English and immediately ask it to switch into Spanish. Notice how it not only changes the language but also changes the accent to a localized accent.

Try it in Anam Lab

The easiest way to get started is to sign up for Anam Lab. Choose a Cara-4 avatar and an ElevenLabs v4 voice, add your persona's instructions, and enable Director Notes to shape the performance.

Try the same passage with different emotional directions, then ask the avatar to switch languages. You can adjust the character and delivery directly in Lab as you go.

Frequently asked questions

What makes an AI voice expressive?

An expressive AI voice varies its delivery to suit the meaning and emotional context of the words. Changes in pacing, emphasis, and tone can make the same sentence feel reassuring or confrontational.

Can a voice keep its identity across languages?

ElevenLabs says Eleven v4 can retain a speaker's vocal identity while adopting a native accent in another supported language. Review the result with speakers of your target language to check pronunciation and delivery.

How does Anam pair expressive speech with an avatar?

Anam's Cara-4 model generates real-time interactive avatars from speech audio, and Director Notes guide the visual performance.

Does Anam's Director Notes feature control exact gestures?

Director Notes guides a Cara-4 avatar's general performance through styles and cues. Director notes guide broad performance rather than exact gestures.