Synthesia has introduced Roleplay Sessions, a platform allowing enterprise employees to practice professional tasks with interactive AI avatars. These digital twins use integrated voice-to-text and video models to listen, respond, and score user performance during simulations like sales pitches.
Interactive training through digital twins
Synthesia has expanded its video-generation technology to include interactive avatars capable of real-time conversation. The new Roleplay Sessions product allows employees to engage in back-and-forth dialogue with a digital twin trained on specific datasets. The system uses a tech stack combining video and language models to process verbal input and generate appropriate visual and vocal responses.
To create these avatars, the company records two minutes of a subject's voice and captures high-resolution imagery. The resulting digital twin can be configured to read scripts or act as an interactive agent. While Synthesia provides its own proprietary video and voice models, the platform supports third-party integrations from providers such as ElevenLabs, OpenAI, and Google.
Deployment and technical constraints
Enterprises can choose to host these avatars on their own cloud infrastructure or utilize Synthesia's hosting services. In recent tests, the interactive capabilities were limited to information contained within specific training documents provided during the setup phase. This ensures the avatar remains focused on the designated subject matter, such as corporate PR or specific articles.
The development represents a transition from one-way video generation to interactive simulation. However, the accuracy of the feedback and scoring system depends on the quality of the underlying language models selected by the enterprise client. Availability is currently focused on corporate training environments rather than general consumer use.
Original source
This report summarises the source below. Analysis is labelled separately; product and research claims remain attributed to their source.
Read the original at TechCrunch AI