Async Flash v1.5 is a high-speed text-to-speech (TTS) model developed by Async (formerly known as Podcastle). It is specifically engineered for high-throughput, ultra-low-latency applications such as real-time conversational AI agents, interactive voice response (IVR) systems, and automated content production. The model is the flagship of Async's "Flash" series, which emphasizes a balance between rapid inference speeds and natural-sounding prosody.
A key technical focus of Flash v1.5 is its sophisticated text normalization engine. The model is designed to handle non-standard text formats—such as complex date structures, currencies, phone numbers, and abbreviations—with higher accuracy than general-purpose TTS systems. It supports instant voice cloning, allowing for the generation of a high-fidelity synthetic voice from a reference audio sample as short as three seconds without requiring specialized fine-tuning.
Capabilities and Performance
Flash v1.5 provides a library of over 500 unique voices and supports more than 15 languages, including major European and Asian variants. In latency benchmarks, the model has demonstrated a median Time to First Byte (TTFB) of approximately 166 milliseconds, making it suitable for streaming environments where immediate responsiveness is critical. It includes support for the International Phonetic Alphabet (IPA) to define custom pronunciations for brand names or technical jargon.
For developers and enterprise users, the model offers integrated timing controls and the ability to insert precise silent pauses using the <break> tag. It has been noted for its price-to-performance efficiency, frequently ranking as a competitive alternative to high-tier proprietary models in independent quality benchmarks. For best results, it is recommended to utilize the model's WebSocket integration to maintain persistent conversation contexts and reduce per-request overhead during interactive sessions.