Inworld Realtime TTS-2

Inworld's Realtime TTS-2 text-to-speech model, first shown in May, is now generally available. Instead of picking from preset emotions, you describe the delivery you want in plain English, and Inworld says the model listens to the conversation audio to match the player's tone. It keeps one voice consistent across more than 100 languages and supports inline cues like [laugh] and [sigh]. Inworld pitches this as a big step up in how directable an NPC voice can be. It also says the model ranks first on Artificial Analysis; that is Inworld's claim. Pricing is metered per character at the same rate as TTS 1.5, so costs scale with how much your characters talk.

Read the official post