Smallest.ai Unveils Lightning V2 Multilingual Text-to-Speech

Foundational AI company Smallest.ai has launched Lightning V2, a text-to-speech (TTS) model with ultra-low latency and deep customization supporting 16 languages across the United States, Europe, the Middle East, and India.

With 100ms streaming latency, voice cloning from just 10 seconds of audio, and seamless language switching, it enables real-time, emotionally rich voice experiences across customer support, virtual assistants, booking systems, and more.

Whether it's English with a Midwest accent, Hindi with natural pauses, or Arabic spoken with regional fluency, Lightning V2 delivers voices that are less synthetic and more realistic. Its voice cloning capabilities enable unique voice identities using a few seconds of sample audio to replicate known voices or create localized personas for different regions.

Beyond voice quality, Lightning V2 is built on Smallest.ai's Full-Stack Voice AI Platform, a system where the AI model, infrastructure, and application layer are developed in house. It learns, adapts, and evolves with usage. Users can switch prompts on the fly as conversations move through stages; tune speech outputs for specific domains like fintech, healthcare, or education; and identify errors in real time and adjust without full model retraining.

The 16 supported languages include U.S., U.K., and Indian English, Hindi, Arabic, German, Tamil, and Telugu and also include fluent handling of numbers, currency, and special characters.

"Conversations aren't static. They have stages, emotions, and shifts in intent," said Sudarshan Kamath co-founder of Smallest.ai, in a statement. "Our platform gives teams the tools to adapt the AI mid-conversation, not just before deployment. That's the difference between good voice automation and great customer experience."