Nuance Labs has introduced Emma, an AI persona designed to make video conversations with artificial intelligence feel more like face-to-face interactions. Launched on November 9, 2026, Emma is powered by the company’s Nuplex 1.0 audiovisual model, which can listen and generate spoken responses while simultaneously producing facial expressions, eye movements and other visual reactions. Unlike conventional voice assistants that generally wait for a person to finish speaking, Emma is designed to react while the conversation is still happening.
The launch also includes the Nuplex Beta API, allowing developers to build applications around AI personas. Nuance says its internal testing found that Nuplex matched human conversational reply timing 89% of the time across 1,785 questions. This is a company-reported result, not an independently verified benchmark, and the technology still has limitations involving interruptions, visual understanding, language support and occasional facial glitches.
Key Takeaways
- Nuance Labs launched Emma on November 9, 2026.
- Emma is powered by Nuplex 1.0, a full-duplex audiovisual AI model.
- The AI persona can listen, speak and display facial reactions simultaneously.
- Nuance says its internal tests matched human reply timing 89% of the time across 1,785 questions.
- Emma’s face and voice are licensed from a real actress who participated in the development process.
- The company has also launched the Nuplex Beta API for developers.
- Developers can create customised AI personas and connect them with external tools.
- Emma is currently English-only and intended for adults aged 18 and above.
- Nuance prohibits unauthorised cloning of real people’s faces and voices.
- The technology still experiences occasional interruptions, visual glitches and mistakes.
What Is Emma, the New AI Persona From Nuance Labs?
Emma is an interactive AI character that users can communicate with through a video-based conversation.
Instead of presenting a conventional chatbot interface with text messages or a voice assistant with a static visual representation, Emma appears as a digital person who responds through speech, facial expressions and body language.
The company says the development involved 18 months of research.
One of its central objectives was to address a problem with existing AI video assistants: many can produce realistic speech and facial movements, but their reactions do not always follow the natural rhythm of human conversations.
For example, when someone speaks to another person, the listener may nod, smile, change their gaze or show an expression before the speaker finishes.
These small reactions communicate attention and help people understand when someone is listening, thinking or preparing to respond.
Emma attempts to reproduce this behaviour.
According to Nuance, the AI persona can:
- Listen while generating its own responses.
- React visually while a person is speaking.
- Use eye movements to communicate attention.
- Display different facial expressions.
- Nod and move its head naturally.
- Maintain a consistent personality during conversations.
- Respond with opinions rather than simply agreeing with everything a user says.
The company describes Emma as an AI persona rather than a digital human intended to deceive people.
It also makes clear that Emma is artificial intelligence and does not claim to be a real person.
What Is Nuplex 1.0?
The technology behind Emma is Nuplex 1.0, which stands for Nuance Full-Duplex.
It is an audiovisual foundation model designed to process incoming audio and visual information while generating its own speech and facial video.
This approach differs from many conventional conversational AI systems.
Traditional voice assistants generally follow a turn-based process:
- The user speaks.
- The system receives and processes the audio.
- The system prepares a response.
- The AI begins speaking.
- The user responds after the AI finishes.
This can create a noticeable delay and make conversations feel less natural.
Nuplex is designed to process listening and speaking simultaneously.
What does full-duplex AI mean?
In telecommunications, full-duplex communication allows information to travel in both directions simultaneously.
Nuance applies a similar concept to conversational AI.
The model can process a person’s voice and camera input while generating its own speech and visual reactions.
This makes it possible for Emma to show a reaction before the user has completely finished speaking.
For example, if someone makes a humorous comment, Emma may smile or react visually while listening rather than waiting until the entire sentence is complete.
Gaze is also treated as a conversational signal, helping the persona communicate when it is listening or preparing to speak.
This is important because natural communication involves more than words. Timing, facial expressions, pauses and visual attention all influence how people understand conversations.
Nuance Claims Emma Matches Human Reply Timing 89% of the Time
One of the most notable claims in the launch announcement concerns conversational timing.
Nuance says it evaluated Nuplex using 1,785 questions taken from its own recordings of real conversations.
The company compared when different AI systems began responding with when the original human speaker responded.
According to its findings, Nuplex’s reply timing overlapped with the original human speaker’s timing 89% of the time.
The company reported that the next closest system reached 35%.
Nuance’s reported conversational timing comparison
| Measurement | Reported result |
|---|---|
| Questions evaluated | 1,785 |
| Nuplex timing overlap | 89% |
| Next closest system | 35% |
| Research duration | 18 months |
The results suggest that Nuance is focusing on a different aspect of conversational AI.
Many AI voice products concentrate on reducing response latency. However, responding as quickly as possible does not necessarily produce a natural conversation.
A person may pause before answering, interrupt briefly or react visually while listening.
Nuance argues that matching these conversational patterns is more important than simply producing the fastest response.
There are important limitations to the comparison.
The benchmark was developed by Nuance using its own recordings and testing methodology. It has not been independently replicated.
The company also acknowledges that there is no established benchmark that fully measures the naturalness of face-to-face audiovisual AI conversations.
Consequently, the 89% figure should be treated as a company-reported research result rather than proof that Emma communicates like a human in every situation.
Emma Has a Licensed Human Face and a Designed Personality
Another interesting aspect of Emma is how Nuance developed her appearance and personality.
Many AI-generated characters use synthetic facial expressions that can appear exaggerated or repetitive.
Nuance says it took a different approach.
Emma’s face and voice are licensed from a real actress who collaborated with the company to capture how she listens, reacts and speaks.
The development team studied specific mannerisms, including:
- Eye movements and blinking.
- Changes in facial expressions.
- Head movements.
- Natural reactions to different emotions.
- Small gestures during conversations.
The objective was to create a consistent character rather than an artificial face that simply changes expressions based on emotion labels.
Why personality matters
Nuance also says Emma has a deliberately designed personality.
The company describes her as curious, quick-witted and capable of disagreeing with users.
This addresses a common criticism of conversational AI: some assistants can be excessively agreeable and repeatedly validate a user’s opinions.
Emma is designed to maintain a more consistent conversational character.
However, her underlying reasoning still comes from a language model. This means she can make factual mistakes or provide confident but incorrect answers.
Her personality should not be confused with human consciousness or genuine emotions.
Nuance Launches Beta API for Developers
Alongside Emma, Nuance Labs has introduced the Nuplex Beta API.
The API allows developers to create applications using AI personas that can communicate through real-time audiovisual conversations.
Developers can provide instructions defining how a persona should behave and interact with users.
They can also connect external tools so that an AI persona can hand longer tasks to those tools while continuing the conversation.
For example, a developer could create an AI assistant that discusses a task with a user, accesses a connected service and works on the request in the background.
The idea is to make AI personas useful beyond simple video conversations.
Potential developer applications
| Industry | Potential application |
|---|---|
| Customer service | Interactive customer-support assistants |
| Sales | Conversational product demonstrations |
| Business | AI assistants for internal workflows |
| Entertainment | Interactive digital characters |
| Research | Conversational interfaces for information access |
| Productivity | AI agents connected to external tools |
These are potential applications of the technology, not a list of fully launched industry-specific products.
Nuance says the beta is intentionally limited while the company continues developing its platform.
The API documentation and commercial terms should be checked before developers build production services around it.
Safety Restrictions and Privacy Controls
Creating AI characters that look and behave like people introduces important safety questions.
A realistic AI persona could potentially be misused for impersonation, manipulation or unauthorised digital likeness creation.
Nuance says it has introduced several restrictions to address these concerns.
Key safety measures
1. Clear AI identification
Emma must always identify herself as artificial intelligence. Developers cannot instruct her to claim that she is a real person.
2. No unauthorised cloning
The Nuplex Beta API does not allow users to upload a person’s face or voice for cloning. Custom faces are available only through arrangements involving explicit permission.
3. Visible watermarking
Video generated through the API carries a visible watermark to help identify AI-generated content.
4. Age restrictions
Emma and the Nuplex Beta API are restricted to adults aged 18 and above.
5. Restricted sensitive applications
Healthcare and applications involving people under 18, including education, are excluded from the beta.
6. Conversation monitoring
Nuance says automated checks can end conversations involving certain prohibited content or dangerous requests. The company also checks persona responses to prevent requests for sensitive information such as payment-card details, passcodes or money.
These restrictions are important because realistic audiovisual AI can create new opportunities for both legitimate communication and digital impersonation.
However, automated safeguards cannot guarantee that every harmful interaction will be detected.
Current Limitations of Emma
Despite its advanced conversational design, Emma is still an early-stage AI product.
Nuance has openly acknowledged several limitations.
| Limitation | Current situation |
|---|---|
| Language support | English only |
| Conversation timing | Can interrupt or respond too late |
| Long pauses | Difficult to handle naturally |
| Visual understanding | Can miss details or misinterpret scenes |
| Facial animation | Occasional lip-sync and gaze glitches |
| Background noise | Can reduce conversational quality |
| Reasoning | Can produce incorrect answers |
| Singing | Not supported |
The company recommends using headphones because background voices, television audio and speakerphone noise can affect conversational timing.
The visual understanding system is also still developing. Emma can discuss what she sees through the camera, but she may misunderstand objects or situations.
These limitations demonstrate that building a convincing audiovisual AI persona involves more than generating a realistic face.
The system must also understand context, manage conversational timing and produce appropriate reactions.
Where Can Users Try Emma?
Nuance says Emma is available through its dedicated website, where users can start a conversation without creating an account.
The company lists availability in:
- The United States, excluding Colorado, Connecticut, Illinois and Texas.
- The European Union.
- The United Kingdom.
- Canada, excluding Quebec.
The service is intended for adults aged 18 and above.
The Nuplex Beta API is available through the company’s developer platform.
Indian availability was not included in the launch’s listed supported regions, so users in India should not assume that the service is currently accessible.
Official platforms:
- Emma: https://emma.nuancelabs.ai/
- Nuance Labs: https://home.nuancelabs.ai/
- Developer platform: https://platform.nuancelabs.ai/
What Emma Means for the Future of AI Video Assistants
The launch highlights a growing shift in conversational AI.
Earlier generations of AI assistants focused mainly on understanding spoken commands and generating voice responses.
Newer systems are attempting to create more natural interactions by combining speech, facial expressions, visual understanding and real-time reactions.
This could change how people interact with digital assistants.
Instead of typing into a chatbot or speaking to a voice-only assistant, users could communicate with an AI character that responds through a combination of speech and visual behaviour.
The technology could eventually support customer service, virtual sales representatives, interactive entertainment and workplace assistants.
However, the market is becoming increasingly competitive.
Companies including Google, OpenAI and other AI developers are investing in real-time voice and audiovisual systems.
Nuance Labs is attempting to differentiate itself by focusing on conversational timing, visual reactions and consistent digital personalities rather than treating the face as a simple animation layered over a voice model.
Its success will depend on whether these capabilities can work reliably at scale and provide enough practical value to developers and businesses.
The Bigger Picture
Emma represents an important development in the evolution of AI assistants, moving beyond text and voice towards more interactive audiovisual communication. By combining simultaneous listening and speaking with facial expressions, gaze and personality, Nuance Labs is attempting to address the subtle behaviours that make human conversations feel natural. The company’s emphasis on conversational timing also suggests that the next generation of AI assistants may compete on interaction quality rather than response speed alone.
However, realistic digital personalities also introduce challenges involving impersonation, privacy, safety and user trust. Nuance has introduced restrictions around identity cloning, age, sensitive applications and AI disclosure, but the effectiveness of these protections will need to be demonstrated over time. Emma is an early step towards more natural AI communication, not yet a replacement for human interaction.
Frequently Asked Questions
1. What is Emma AI?
Emma is an AI persona developed by Nuance Labs. It allows users to have audiovisual conversations with a digital character that speaks, listens and displays facial reactions in real time.
2. What is Nuplex 1.0?
Nuplex 1.0 is Nuance Labs’ full-duplex audiovisual AI model. It processes incoming audio and visual information while generating speech and facial video simultaneously.
3. Is Emma a real human?
No. Emma is an artificial intelligence persona. Her face and voice are licensed from a real actress who collaborated with Nuance Labs during development.
4. What does full-duplex AI mean?
Full-duplex AI can listen and generate responses simultaneously rather than waiting for a user to finish speaking before beginning its own response.
5. How accurate is Emma’s conversational timing?
Nuance reports that its model achieved 89% timing overlap with human responses across 1,785 questions from its own conversation recordings. This result has not been independently verified.
6. Can developers build their own AI personas using Nuplex?
Yes. Nuance has launched a Beta API that allows developers to customise personas through instructions and connect them to external tools.
7. Is Emma available in India?
India was not included in the supported regions listed in Nuance’s launch announcement. Availability should be checked on the official Emma website.
8. Can Emma clone a real person’s face or voice?
No. Nuance says the beta API does not support unauthorised cloning or uploading another person’s face and voice. Custom likenesses require explicit permission.
9. Is Emma free?
Nuance’s official launch announcement provides access to Emma but does not establish a detailed commercial pricing structure for the service. Developers should check the official platform for current API terms.
Looking Ahead
Nuance Labs’ launch of Emma and the Nuplex Beta API could help push conversational AI towards more natural video-based interactions. If the company can improve visual understanding, reduce interruptions and maintain consistent facial behaviour, its technology could become useful for customer engagement, interactive assistants and digital services. The developer API is particularly important because it gives businesses an opportunity to experiment with audiovisual AI personas beyond the company’s own Emma experience.
The bigger challenge will be turning an impressive demonstration into a reliable product that businesses and consumers can trust. Independent performance testing, transparent pricing, wider language support and stronger privacy protections will all influence adoption. Emma demonstrates how quickly AI interaction is evolving, but the long-term value of the technology will depend on how naturally, safely and consistently it performs in everyday conversations.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
