This new AI model hears your tone, senses your mood, and talks back like a real human. Siri could never.

Inworld AI, a startup based in Mountain View, has introduced a groundbreaking voice model that aims to make interactions with machines feel more authentic by understanding both the content and emotional nuances of speech. The new system, known as Realtime TTS-2, leverages advanced analysis of vocal characteristics—such as tone, pacing, and pitch—to gauge a speaker's emotional state in real-time, allowing it to adjust its responses to foster more natural conversations. As AI voice technology continues to evolve, the potential for increased user engagement is significant. While previous models have excelled in text-based communication and image generation, the ability to converse with AI in a more lifelike manner is seen as the next frontier. According to Kylan Gibbs, CEO of Inworld, addressing the emotional dimension in AI interactions is crucial for widespread adoption. "People naturally interact through real-time conversation, and enhancing that experience leads to greater engagement," Gibbs shared in a recent interview. This release signifies a strategic shift for Inworld, which has successfully secured over $100 million in funding from prominent investors such as Founders Fund, Intel, and Microsoft. The company previously boasted a leading voice model that outperformed competitors like Google and ElevenLabs, but Gibbs emphasized the need for more than just high-quality voice output. Most existing AI voice models have been tailored for specific applications like audiobooks and voiceovers, but Gibbs identified a key limitation: while these voices sound human, they often come across as robotic and scripted. Inworld's goal with TTS-2 is to bridge the gap between realism and genuine interaction, addressing the disconnect that many users experience. TTS-2 stands out by integrating capabilities that are rarely combined in AI voice systems. For example, it takes into account the full context of a conversation, allowing it to tailor responses based on prior exchanges. The model can detect emotional signals in speech and continuously updates its understanding of both the user and the agent's emotional states, refining its responses accordingly. During an exclusive live demonstration at Inworld's headquarters, Gibbs showcased TTS-2's versatility. The model adeptly shifted its tone in response to various contexts, demonstrating an empathetic and direct approach when addressing service delays and a warm demeanor in more casual exchanges. This adaptability was further highlighted by an AI character named "Jason," who provided nuanced responses that conveyed both amusement and polite disapproval following an inappropriate joke. Gibbs noted that emotional awareness has largely been absent in current voice AI technologies, which typically treat speech as isolated text inputs. In contrast, TTS-2 interprets a wider range of communicative signals, focusing on delivery style and prosody—essentially how something is conveyed rather than just the words chosen. The implications of this technology are vast, encompassing fields ranging from customer service and healthcare to education and digital companions. Inworld plans to offer TTS-2 as an infrastructure tool for developers, making it accessible via an API that integrates with existing AI systems. This approach allows developers to create tailored applications without competing with Inworld's own offerings. With the rise of AI coding tools simplifying app development, Gibbs believes there is less emphasis on the application layer of technology. Instead, Inworld is dedicated to providing robust models and APIs, positioning itself as a key player in the evolving landscape of AI-driven communication.