Inworld Realtime TTS-2 adds voice direction

This looks less like a cosmetic TTS refresh and more like a real step toward promptable voice infrastructure for agents. The pitch is simple: if voice is going to be an interface, developers need something they can steer as precisely as an LLM.

• –Natural-language direction lowers the friction between product intent and audio output, which matters more than raw fidelity in most agent workflows
• –Text-based voice design is useful for teams that want to prototype brand voices without recording sessions or casting
• –Cross-lingual synthesis with identity preservation is the strongest differentiator for globally deployed support, companion, and tutor apps
• –IPA control and better alphanumeric pronunciation target the trust breakers that still make voice products feel brittle in production
• –This is a strong fit for AI agent stacks, but it is still a specialized TTS product rather than a broad platform move