AssemblyAI has released a Voice Agent API designed for low-latency speech-to-speech interactions, charging a flat $4.50 per hour of audio processed. The offering targets developers building real-time AI voice applications, from virtual assistants to conversational interfaces.
What the API does
The API handles the full speech pipeline — from capturing a user's spoken input to generating a spoken response — in a single, integrated call. AssemblyAI says its system is optimized for speed, reducing the lag that often plagues voice-based AI interactions. For developers, that means they can skip stitching together separate speech recognition, natural language processing, and text-to-speech services.
Pricing at a flat rate
Instead of charging per request or per character, AssemblyAI is offering a simple hourly rate. The $4.50 per hour covers the entire speech-to-speech conversion, with no additional fees for the underlying models. Company representatives say the flat pricing is meant to make costs predictable for startups and enterprises alike.
Who's the target
The API is aimed at any application where a user talks to an AI and expects a near-instant spoken reply. Think customer support bots, voice-controlled apps, or even in-car assistants. Real-time voice use cases have grown rapidly, but many existing solutions still rely on multi-step processing that introduces noticeable delays. AssemblyAI's pitch is that its API cuts that delay to a minimum.
Developers can begin using the Voice Agent API immediately through AssemblyAI's existing developer portal. The company is also offering a free tier with limited hours for testing.
