Inworld acquired Ultravox and now controls both ends of the voice AI stack.
Ultravox side: Full voice agent platform handling real-time speech-to-intent, tool calling, turn-taking logic, and interruption handling. Core team joining Inworld to continue development.
Inworld side: Realtime TTS-2 model ranked #1 on Artificial Analysis. Sub-100ms latency. 200+ language support. Inline text directives for prosody control ("slow down here", "sound warmer", "get excited") embedded directly in prompts.
Technical implications: Single vendor now owns both the conversational reasoning layer (when to speak, what actions to trigger) and the voice synthesis layer (how it sounds, latency, multilingual rendering). Existing Ultravox deployments auto-upgrade to TTS-2 with zero migration work.
Use cases: AI companions, tutors, customer service agents where user retention hinges on conversational fluidity and human-like interaction patterns. The bet is vertical integration reduces latency and improves coherence between intent understanding and voice output timing.
Ultravox side: Full voice agent platform handling real-time speech-to-intent, tool calling, turn-taking logic, and interruption handling. Core team joining Inworld to continue development.
Inworld side: Realtime TTS-2 model ranked #1 on Artificial Analysis. Sub-100ms latency. 200+ language support. Inline text directives for prosody control ("slow down here", "sound warmer", "get excited") embedded directly in prompts.
Technical implications: Single vendor now owns both the conversational reasoning layer (when to speak, what actions to trigger) and the voice synthesis layer (how it sounds, latency, multilingual rendering). Existing Ultravox deployments auto-upgrade to TTS-2 with zero migration work.
Use cases: AI companions, tutors, customer service agents where user retention hinges on conversational fluidity and human-like interaction patterns. The bet is vertical integration reduces latency and improves coherence between intent understanding and voice output timing.
