“What is the refund window if I already shipped it back?”Speech to text · English
Speech becomes text. The language can change. The model answers from your sources. The reply comes back as audio, or as a talking-avatar video. The same agent can take the next step in the app.
The screens above are a sample of the workspace. They play on their own. Select a step to hold it.
A support line, a kiosk, or a cabin cannot hand someone a document. Typing is the wrong interface. A text-only bot leaves the spoken question unanswered in the language it arrived in.
EazyAI transcribes, translates when needed, answers from the project, and speaks back. You can return audio or a video of the avatar. Citations still sit with the answer for the team that reviews it.
Four moves. The sources are yours. The answer names them. A person keeps the decision that matters.
A wake word can open the mic. Speech becomes text on your setup.
The reply is prepared in the language that was spoken.
The same grounded answer as the text agent, with the source kept for review.
Audio, or a talking-avatar video. A confirmation can gate the next action in the app.
The working shape of this use case. Live in two to three days on the sources you already have.
The caller or the visitor talks. The transcript is the question.
Translation sits between the transcript and the reply when it is needed.
The same answer can be spoken or shown as a talking-avatar video.
The project runs on your premises or your private cloud. Dual RAG cites the passage. The model you serve is the one you register. A demo is the way onto a first workflow.
STT → translate → LLM → audio or talking-avatar video. Confirm-before-act where policy requires it. Works with embed and cabin edge.
Wake word, duplex voice, Dual RAG, then Kokoro or Piper — or a talking-avatar reply. Deep links and approved agents run the next step after a spoken yes when you require it. Offline cabin modules use the same pattern on the aircraft.
Reply in the language that was spoken.
Sensitive actions wait for a spoken yes.
Same voice pattern online or offline on a seatback stack. See Offline Voice.
What product and cabin leadership ask about voice.
On hosts you allowlist. Endpoints stay on your stack.
Both. Kokoro/Piper for audio; optional talking-avatar video from the same turn.
Offline modules run on-device after the pack is loaded. Sync when a link returns. See aviation and Offline Voice on the home page.
Same platform. A different set of sources and a different person in the loop.
Book a demo. A first workflow in two to three days. Your data stays where it is.