Introducing AssemblyAI-TTS-1: text-to-speech built for voice agents

AssemblyAI announced the general availability of AssemblyAI-TTS-1, a text-to-speech model designed for voice agents that it prices at $15 per million input characters and says streams first production audio in about 90 milliseconds, with a 47 ms median time to first byte in its internal API tests. The company positions the model as a response to call abandonment caused by slow or robotic automated systems, claiming it uses continuous audio representations trained on millions of hours of audio to improve naturalness and reduce latency. Features include more than 4,000 voices, instant consent-based voice cloning, word timestamps, barge-in cancellation support, unlimited pay-as-you-go concurrency, and US or EU data processing, with English, Spanish, French, German, Italian, and Portuguese available at launch. In comparisons conducted by AssemblyAI against several competing models, TTS-1 reportedly achieved 99.0% word accuracy on standard sentence sets, 93.7% entity accuracy on difficult pronunciations, and competitive automated naturalness scores; the company also cites results from independent benchmarks. At launch, dedicated controls for pronunciation, emotion, speed, SSML, and fine-tuned “Pro” voice cloning are not yet available, though AssemblyAI says several are planned.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.