AssemblyAI launches AssemblyAI-TTS-1, its first text-to-speech model for voice agents
IA en un minuto newsroom · Editor: Jon Elgezabal
In 30 seconds
AssemblyAI has released a debut speech-synthesis model, which turns the written replies of an agent answering phone calls into audio. It is billed per character, with no annual contract and no cap on simultaneous conversations, and comes with a library of over 4,000 voices plus cloning of a custom one. It works in six languages, Spanish among them, and the company says speech starts in under 100 milliseconds.
AssemblyAI, known for its speech-to-text models, has launched AssemblyAI-TTS-1, its first text-to-speech model: the one that turns an assistant's written replies into spoken audio. According to its blog, it is built for voice agents that handle phone calls and has been open to everyone since October 27, 2026.
Pricing is $15 per million characters, with no spending commitment or annual contract. The company says there is no cap on simultaneous calls: the same per-character price applies to ten conversations as to ten thousand. It also presents its rate as the lowest among top-tier voice models and as roughly half of what, it says, is paid today.
It includes a library of more than 4,000 voices and instant cloning: a recording is uploaded, the speaker's permission is confirmed, and the voice is ready within seconds. It works in six languages, with several accents in English, Spanish, French, German, Italian and Portuguese, and AssemblyAI plans to add more languages and accents in the coming weeks. For now it does not offer fine-tuned "Pro" cloning, and pronunciation dictionaries and emotion tags will come later.
On speed, the company says the voice starts in under 100 ms and puts time to first audio in production at 90 ms. In its lab measurement, the median time to first byte was 47 ms, against 119 ms for ElevenLabs Flash v2.5, the next fastest of the six models compared. AssemblyAI ran those tests itself, from a single client and with English sentences; they do not come from an independent evaluation.
Why it matters · analysis and opinion
The launch says something about the voice agent market: a company that already sold the part about understanding the customer on the phone now adds the part about speaking to them, and competes on a per-character price with no annual commitment. For a business thinking of handling calls with an agent, the voice bill grows with every conversation, so price and the lack of strings weigh almost as much as sound quality. The comparisons the company cites are its own, run from a single client and with English sentences. The prudent step is to test the service with real calls in Spanish and measure how many are resolved without passing to a person before deciding.
Official source: AssemblyAI · Written with the help of AI: how we make the news
