Inworld TTS
Inworld Realtime TTS is a real-time text-to-speech technology from Inworld that provides voice models designed for applications requiring fast, realistic, and interactive speech generation. The platform includes Realtime TTS-2, the current flagship model, and Realtime TTS-2 Flash, designed for applications requiring low latency and lower cost, as well as previous Realtime TTS models such as Realtime TTS 1 and Realtime TTS 1-Max. Realtime TTS-2 offers advanced vocal expression and the ability to control delivery style using natural language instructions; it also supports nonverbal cues such as laughter, sighs, breathing, coughing, yawning, and throat-clearing, and a single voice can be used across multiple languages. The platform offers instant voice cloning using a short voice sample, as well as Voice Design for creating custom voices. It also supports control over speech rate, pitch, custom pronunciation, timestamps, and alignment data. Realtime TTS-2 offers a latency of less than 100 ms based on server-side TTFB measurements, while Realtime TTS-2 Flash reaches 25 ms, according to the official website. The current models also support more than 200 languages with multilingual capabilities and language switching using BCP-47 codes.