Async Pro v1.0 is a high-fidelity text-to-speech (TTS) model developed by Async, a voice AI platform formerly known as Podcastle. As the premium tier in the company's model lineup, it is designed for production-grade applications that require natural prosody and precise articulation. The model distinguishes itself through a focus on intelligent speech synthesis, prioritizing the correct handling of complex text elements that often cause failures in standard streaming systems.
The model's primary technical strength lies in its text normalization capabilities. It is specifically engineered to accurately pronounce non-standard tokens such as dates, currencies, phone numbers, and technical abbreviations without the need for manual text preprocessing. In industry benchmarks, Async Pro v1.0 has been noted for its ability to maintain high sentence-level accuracy in real-time streaming environments, where traditional normalization layers might introduce significant latency or errors.
Beyond standard synthesis, Async Pro v1.0 supports instant voice cloning, capable of generating a production-ready voice replica from a three-second audio sample. It provides developers with granular control via Custom Phonemes (using IPA), digit-by-digit pronunciation toggles, and adjustable silent pauses using specialized tags. These features allow for the customization of brand-specific terminology and the fine-tuning of speech rhythm for more natural human-computer interaction.
Architecturally, the model is optimized for low-latency delivery, supporting first-byte audio generation in approximately 300 to 400 milliseconds. It is primarily utilized via WebSocket or HTTP streaming protocols, allowing for the concurrent generation and playback of audio. While the specific parameter count is proprietary, the model is recognized in the voice AI industry for its balance of high-quality vocal output and computational efficiency.