Ultravox

Ultravox.ai is a real time speech AI infrastructure layer. It is built to fix common problems in traditional speech systems, such as the delay caused by first turning speech into text and the loss of paralinguistic cues like tone and emotion. The page lists low latency, keeping those vocal cues, and fast, natural, scalable speech interaction as its main advantages. It comes from a research lab and product company that trains its own speech models and runs them on dedicated infrastructure. Pricing includes a free trial described as 5 cents per minute with up to 5 concurrent calls, a pay as you go option with concurrency limits, a professional version at $100 per year, and a custom enterprise plan. It is aimed at developers and enterprises.

- Build voice agents that respond in real time with low delay

- Keep vocal cues like tone and emotion in the interaction

- Scale speech features for a product or service

- Start on a small free trial before moving to paid usage

- Developers adding speech to an application

- Enterprises building conversational voice experiences

- Teams that need scalable, low latency voice infrastructure

The free trial is capped, at 5 cents per minute and up to 5 concurrent calls, and pay as you go has concurrency limits, so costs rise as usage grows and free trial terms can change. If you build voice agents for outbound or automated calling, you must follow the call, consent, and privacy rules that apply in your region, and any misuse is your responsibility. Because it is infrastructure, you will need engineering work to integrate it, and it relies on the provider's models staying available and accurate. Plan for ongoing monitoring of latency and cost as you scale.