Inworld AI Unveils Cut-Price Open-Source TTS Model for Real-Time Voice Apps

Inworld AI released a new generation text-to-speech system, TTS-1, that the company says cuts the cost of high-quality voice synthesis to US$5 per million characters—about one-twentieth the prevailing market rate. The model, available immediately via API and in a browser-based playground, is designed for real-time applications such as gaming avatars, virtual assistants and fitness trainers, delivering the first two seconds of audio in as little as 200 milliseconds. The launch includes an experimental, more expressive variant dubbed TTS-1-Max and free zero-shot voice cloning that can replicate a speaker from a brief sample. Inworld is also opening its Llama-based training and inference code under a commercially permissive licence, pledging to publish a detailed technical report in the coming weeks. The company says the model supports 11 languages and embeds an imperceptible watermark to flag AI-generated audio. Safeguards are in place to prevent unauthorized voice cloning, and further capabilities—such as generating voices from text descriptions—are under development.