AssemblyAI Launches Dictation API That Fills Whole Forms by Voice

AssemblyAI, a speech-recognition company, launched its Dictation API with a post aimed at a familiar annoyance: forms that spread questions across many pages and fields. The pitch is to answer all of it with one spoken prompt. The post claims dictation lets users enter text up to three times faster; it gives no method behind that figure, so it is the company's claim, not a measured result.

The demo video shows the idea on a patient intake form. The presenter talks through one recording covering his name, date of birth, the medicines he takes, how long he has had pain, allergies and his pain level. When he stops, the page reports that six of the ten questions are filled, with each answer in its field, and he says the rest can be answered by voice the same way. Only an automatic transcript of the video was available, so the screen wasn't reviewed, and the material doesn't say whether the API itself maps speech to form fields or the demo app does that on top of it.

For developers, the video points to AssemblyAI's dashboard playground. There you can list key terms, such as technical, legal or medical vocabulary the transcription should get right, and write a prompt describing the output you want, such as an email or bullet points. AssemblyAI says the API runs on its Universal-3.5 Pro speech model. A follow-up post explains the other half of the design: one call both transcribes the speech and turns it into finished text.