Async Flash v1.0 is a low-latency text-to-speech (TTS) model developed by Async for real-time conversational AI and voice agent applications. Formerly known as AsyncFlow 1.0, the model is architected to prioritize speed and responsiveness, achieving a median time-to-first-byte (TTFB) of approximately 0.166 seconds. It is designed to facilitate fluid human-AI interaction by reducing the delays typically associated with traditional speech synthesis pipelines.
The model supports synthesis in over 15 languages and features instant voice cloning capabilities, which can generate a neural clone from a three-second audio reference. Users can further customize output through style, pace, and pronunciation controls to maintain consistent brand voices across global markets. Async Flash v1.0 is frequently utilized in scenarios where low latency is critical, such as customer support bots, interactive NPCs, and live streaming tools.
While optimized for speed, the model maintains a high standard of naturalness and prosody, consistently ranking among the top performers on competitive benchmarks like the Hugging Face TTS Arena. It is offered as a developer-first API, providing both REST and WebSocket endpoints for seamless integration into existing software stacks.