Tavus Introduces Raven-1, Bringing Multimodal Perception to Real-Time Conversational AI
Tavus, a company specializing in human computing, has launched Raven-1, a multimodal perception system designed to enhance real-time conversational AI. Raven-1 enables AI to understand emotion, intent, and context similarly to human perception.
Tavus, a company specializing in human computing, has launched Raven-1, a multimodal perception system designed to enhance real-time conversational AI. Raven-1 enables AI to understand emotion, intent, and context similarly to human perception.
Raven-1 integrates audio and visual signals, allowing AI systems to comprehend not just the content of user communications but also the underlying tone and intent. This capability is now available across all Tavus conversations and APIs.
Audio-visual fusion that combines tone, prosody, facial expression, posture, and gaze into a unified real-time context. Natural language outputs directly aligned with large language models (LLMs), eliminating the need for a translation layer. Temporal modeling to track the evolution of emotional and attentional states throughout a conversation. Sub-100ms audio perception latency with a combined pipeline latency of under 600ms. Custom tool calling support for developer-defined events, such as emotional thresholds and attention shifts.
Raven-1 works as a perception layer alongside other Tavus models, such as Sparrow-1 and Phoenix-4, forming a closed loop where perception informs response and vice versa. This development is critical for high-stakes applications such as healthcare, therapy, and interviews, where nuanced understanding is essential.
Tavus, a company specializing in human computing, has launched Raven-1, a multimodal perception system designed to enhance real-time conversational AI.
Designed for real-time operation, Raven-1 excels in processing short, ambiguous, and emotionally loaded inputs. It captures signals that traditional systems might miss, providing valuable context for AI-generated responses.
Raven-1 is now generally available for all Tavus conversations and APIs. The model functions automatically and provides perception layer access through Tavus APIs for custom tool calls and programmatic logic.
For more information, visit the demo at https://raven.tavuslabs.org .
Tavus is a San Francisco-based AI research company focused on human computing, developing models that enable machines to interact naturally with people. The company offers APIs for developers and businesses, as well as PALs, a consumer platform for AI agents.
Based on reporting by TechBullion.