About IVR Voice Gen
IVR Voice Gen generates spoken prompts for phone systems and voice agents. It exists because general-purpose text-to-speech is optimised for prose, and phone systems mostly do not read prose.
What is the problem?
A phone system spends most of its time reading back things that are not words: booking references, callback numbers, verification codes, delivery windows, email addresses. General-purpose models normalise digit strings as quantities before synthesis, so a reference ending in 1222 comes out as "one thousand two hundred twenty-two". That is correct arithmetic and a failed call, because the caller has no screen to fall back on.
Published word error rates do not capture this. They are measured on ordinary transcripts, where these strings barely appear. A model can score well on the benchmark and still be unusable on a phone line.
What is it built on?
IVR Voice Gen is built on Gradium TTS. Latency and accuracy figures for the underlying engine are published on the home page with their source and the date they were read, because that benchmark is a live board rather than a fixed report and the numbers move.
What stage is it at?
Early access. There is a waiting list, not a signup, and no paid plan yet. Five languages are planned for the first release: English, French, German, Spanish and Portuguese, each with its own number, date and currency conventions applied automatically.
What is deliberately not here?
No analytics, no tracking pixels and no cookies, which is why this site has never asked you to consent to any. The privacy page lists the four fields the waiting list stores and nothing is collected beyond them.