LMNT

LMNT is a developer-focused text-to-speech API built for ultra low-latency, real-time streaming voice, with reported latency of around 150-200 ms. It supports fast voice cloning from a short recording and offers polished voices across roughly 24 languages. Designed to scale, with concurrency/rate-limit removal for enterprise, it is aimed at conversational AI, voice agents, and interactive applications. Pricing is freemium and usage-based, from a free tier and an affordable entry plan to per-character overage and custom enterprise rates.

LMNT provides a text-to-speech API engineered for speed, delivering lifelike streaming audio with latency reported in the 150-200 millisecond range. That responsiveness makes it well-suited to conversational AI, voice agents, and interactive experiences where the gap between text and audio must feel instantaneous. It also supports fast voice cloning from a short recording and offers a catalog of polished, natural-sounding voices across roughly two dozen languages. The product is aimed squarely at developers building real-time voice features, with an API designed to scale and, for enterprise customers, options that remove concurrency and rate limits. Pricing follows a freemium, usage-based model with a free tier, an affordable entry plan, per-character overage rates, and custom enterprise pricing. Founded in the late 2010s and based in San Francisco, LMNT is a smaller, focused player in the TTS space rather than a broad consumer brand. Public funding details are limited and somewhat inconsistent across sources, so treat totals cautiously. LMNT is best for teams that specifically need low-latency streaming voice and are comfortable integrating an API.