AssemblyAI-TTS-1

AssemblyAI-TTS-1 is a text-to-speech model developed by AssemblyAI and released on October 28, 2026, specifically engineered for real-time conversational voice agents. Trained on extensive collections of audio data including natural conversational speech, the model is designed to handle live calls with minimal latency, achieving audio generation times under 100 milliseconds to the first audio packet.

Capabilities and Architecture

The model is structured around a streaming text-to-speech framework that accepts text incrementally as an LLM writes it, synthesizing and streaming audio back over a persistent WebSocket session. This allows audio playback to begin long before a full sentence or paragraph is completed. AssemblyAI-TTS-1 supports natural expression, high-concurrency environments without capacity contracts, and provides multiple built-in voices across supported languages.