Mist V2 is a conversational text-to-speech (TTS) engine developed by Rime, engineered for high-performance enterprise applications and real-time voice interfaces. Unlike Rime’s more expressive Arcana model, Mist V2 is optimized for speed, precision, and high-throughput production environments. It is designed to handle complex proper nouns, medical jargon, and brand names with high fidelity, making it a primary choice for IVR systems and business-critical telephony.
The model distinguishes itself through its deterministic pronunciation control, allowing developers to specify exact phonetic renderings via API or dashboard tools. This feature ensures consistent pronunciation of domain-specific terminology across all voices without requiring a manual model retraining cycle. Mist V2 is capable of ultra-fast performance, achieving on-premise latencies of approximately 70ms, which is critical for maintaining natural conversational flows in live AI-driven interactions.
Mist V2 supports multilingual synthesis, with core support for English and Spanish and additional languages added in subsequent updates. It was trained on a proprietary dataset of everyday conversational speech, enabling it to handle natural dialogue patterns, including filler words and breathing, more effectively than traditional neutral-assistant TTS models. The model is frequently cited for its low Word Error Rate (WER) and its ability to maintain high intelligibility across diverse demographics.
For optimal use, the model supports fine-grained control over prosody and pacing. While it shares several voice identities with the Arcana family, Mist V2 prioritizes clarity and predictable delivery. It provides a curated portfolio of production-stable voices, each designed with a distinct tonal identity to suit varied professional and commercial use cases.