Inworld Realtime TTS-2
last releaseMay 5, 2026
powered byinworld-tts-2
goblin vibe check:
the one to watch if your characters need to sound alive instead of reading lines off a spreadsheet
realtime conversational voice model that conditions on prior audio context, voice direction, and user state. useful for game characters, companions, support agents, and multilingual voice interfaces that need more than static narration.
speed
<200ms
first chunk
conditions on the prior audio of the exchange rather than isolated text-only synthesistakes natural-language voice direction inline like a promptpreserves one voice identity across over 100 languagesavailable through both the inworld api and the inworld realtime api
key features
conditions on the prior audio of the exchange rather than isolated text-only synthesistakes natural-language voice direction inline like a promptpreserves one voice identity across over 100 languagesavailable through both the inworld api and the inworld realtime api
spec & usage
official research preview post is dated may 5, 2026
docs claim sub-200ms first-chunk latency for streaming
customers on realtime tts 1.5 upgrade by changing the model identifier
limitations
launched as a research preview rather than a fully settled ga product
long-tail language support is described as launch-window experimental
scope:
audiomodelvoiceapicloudpaidreal-timemultimodal
launchMay 5, 2026
last releaseMay 5, 2026