What happens when you talk to AI?
I made this walkthrough to show how much goes into a simple spoken request: recognizing words, knowing when to reply, and helping someone feel understood. I love thinking about where interface design and engineering meet. Follow a restaurant reservation to hear and see those moving parts in action.
Your voice is the input.
You ask for a table in your own words. The waveform traces the recorded sound over time; speech recognition turns that sound into text. Using familiar language applies Nielsen’s principle of matching the system to the real world: you can ask as you would ask a person.
A pause isn’t the end.
“A table for two…” You think. “…at seven.” Good conversational timing leaves room for that pause. Turn detection uses clues from speech and silence to decide when you’ve finished, taking the place of a Send button.
Make the wait understandable.
“Let me check” tells you your request was heard and something is happening. This is visibility of system status: spoken feedback does the job of a loading indicator. Meanwhile, the language model requests availability from the reservation system, so the answer comes from a lookup rather than a guess.
2 people · 7:00 pm
Checking availability…
The answer becomes speech.
“Seven is available.” The useful answer comes first. Applying aesthetic and minimalist design to voice means keeping replies focused, because listeners must wait through every word. Speech synthesis turns the reply text into sound.
You can change your mind.
“Actually, make that three.” Being able to revise your request is user control and freedom, just like editing a form. Behind the scenes, interruption handling stops playback so the assistant can listen, update the party size, and check availability again.
Seven is available. Would you like me…
Actually, make that three.
Sure, no problem. Let me check availability for three.
The interface changes.
The design questions remain.
A loading indicator and “Let me check” do the same job: tell you the system is working on your request. This is visibility of system status. On a screen, you see the feedback. In a conversation, you hear it.
“A table for two at seven.”
“Let me check.”