Speak to an agent in any language
What it does
Voice input was rebuilt on AssemblyAI's real-time Whisper stream with language detection on, so it transcribes whatever language it hears — including a switch mid-sentence — without anyone choosing one first.
Before this
You could already dictate, but on the browser's own speech recognition: one language at a time, support that varied between browsers, and a stream that stopped at the first pause.
Why it matters
A dictation box that assumes English is unusable for most of the people a studio serves, and picking a language before you speak is a step nobody remembers to take.
How it works
The microphone is where it always was; what changed is behind it. Speech now streams to AssemblyAI rather than to whatever the browser happened to implement, which is also why it behaves the same everywhere.
- Language
- Detected automatically, per turn. Switching language mid-sentence is handled rather than garbled.
- Pauses
- Recording stays live through a pause and across finished turns until you stop it, instead of ending at the first silence.
- Where
- The builder's chat and form inputs, the first-agent prompt at signup, and the feedback widget in both the Builder and the dashboard.
Voice output changed over the same stretch, but that half is the voice picker from 12 August rather than anything new here.