#Voice — Speaking & Listening

Why type when you can talk? And why read when the AI can read to you? Your hands are busy holding groceries, stirring risotto, or just being lazy. We don't judge.

#What It Does

Voice in ambientChat works in both directions. You can speak your questions instead of typing them, and the AI can read its answers aloud instead of making you squint at text. It turns your AI assistant into something closer to an actual conversation — one where the other party happens to know the entire contents of your pantry.

Voice Input uses speech recognition to transcribe what you say into a message for the AI. You speak, it types, the AI answers.

Voice Output uses text-to-speech to read the AI's responses aloud. The same voice, style, and speed you configure once is used everywhere the app speaks: chat auto-play, the speaker icon on individual chat messages, Voice Conversation mode, beacon-triggered room narration, and the voice preview in Settings. Change any of it in one place and it applies everywhere, immediately.

Voice Conversation is a dedicated hands-free mode. Tap the mic in Chat to open a full-screen overlay that listens, thinks, and speaks back continuously — a proper back-and-forth, no tapping required between turns.

#How to Use It

#iOS App

#Speaking to the AI (voice input)

Open the Chat tab
Tap the microphone button next to the text input — or the larger mic to enter full Voice Conversation mode
Wait for the soft "ready" chime — that's the mic going live. Speaking before the chime can clip the first word.
Speak your question clearly
The app transcribes and sends automatically; the AI answers

You can speak naturally. "What's expiring this week?" works as well spoken as typed. The AI doesn't care about your accent — it's heard worse.

#Listening to the AI (voice output)

Every AI message has a small speaker icon on it. Tap to hear the response. Tap again (on any message, or the same one) to stop playback — the app treats the speaker icon as a universal mute. While the audio is being generated, a three-dot throbber appears next to the speaker so you know something's coming during the brief OpenAI round-trip.

With Auto-Speak on, every AI response plays automatically as it arrives. With it off, tap the speaker on any message you want to hear. In Voice Conversation mode, responses auto-play regardless of the Auto-Speak setting — that mode wouldn't be much of a conversation otherwise.

#Voice Settings

Account → Voice Settings. You configure four things:

Voice — choose from six OpenAI voices: Ash, Fable, Marin, Onyx, Sage, Verse. Tapping a voice auditions it immediately so you hear what you're picking.
Voice Style — a free-text field for how the voice should sound. Examples: "Speak warmly like a museum guide", "Sound excited about recipes", "Deep, calm, reassuring tone". Accents are hit-or-miss (the model tries, but won't always land a convincing one). After you tap out of the field, the app auto-auditions so you immediately hear the effect.
Speed — how fast the voice speaks, in seven discrete steps: 0.5×, 0.75×, 1×, 1.25×, 1.5×, 1.75×, 2×. Tapping a chip auditions at the new rate so the change is audible right away. Unlike accents, speed is exact — OpenAI honours it deterministically.
Auto-Speak — toggle automatic playback of every AI response in the main Chat view.

Settings apply across every place the app speaks. Change it once.

#Changing voice mid-conversation (without opening Settings)

In Voice Conversation, just ask. Natural language works:

"Switch to Onyx."
"Let's try Sage."
"Use a warmer voice."
"Can you speak at half speed?"
"Speed it up to 1.5."

The AI updates your saved settings and the next response uses the new voice / style / speed. The Settings screen reflects the change the next time you open it.

#Web

Voice input and spoken replies are an iOS feature. The web interface chat is text-only — you type your questions and read the answers.

#Via Chat

Voice input and text input produce identical results. The AI doesn't know or care whether you typed or spoke — it receives the same text either way. So every example prompt in every other guide in this knowledge base works with voice too. Just say it instead of typing it.

Some particularly useful voice scenarios:

"What's expiring this week?" (while unpacking groceries)
"What spices do I have that go with salmon?" (while staring at salmon)
"Read me the warranty details for my laptop." (while on hold with tech support)
"Add a note: I put the Christmas decorations in the garage attic." (while climbing down a ladder)

#Via Claude Desktop (MCP)

Voice preferences are readable and writable via MCP:

get_voice_preferences — Get your current voice, style, auto-speak, and speed settings
update_voice_preferences — Change voice, style, speed, or auto-speak toggle

For the full tool reference, see MCP Tools — Complete Reference.

#Tips & Tricks

Wait for the chime. It's the honest "mic is hot" signal; speaking before it can clip your first word.
Voice output is great for hands-free workflows. Cooking, organizing the garage, or sorting through boxes — any time your hands are busy but your ears are free.
Tap speaker to stop. Tap any speaker icon to silence whatever's playing. No need to wait out a long response politely.
Change your mind about a voice? Tap the speaker on an older message — the app regenerates audio in your current voice, style, and speed on the fly. Previously-spoken messages keep their old recording until you ask for a fresh one.
Use Speed instead of Voice Style for rate. Typing "speak quickly" in Voice Style is ignored by the model — the model treats that as tonal guidance, not a rate dimension. The Speed chips are the reliable way to make the voice faster or slower.

#Options

Setting	What It Does	Values	Default
Voice	Which OpenAI voice reads responses	Ash · Fable · Marin · Onyx · Sage · Verse	Sage
Voice Style	Free-text instructions shaping delivery	Any short phrase	(empty)
Speed	Playback rate multiplier	0.5× · 0.75× · 1× · 1.25× · 1.5× · 1.75× · 2×	1×
Auto-Speak	Play AI responses automatically in Chat view	On / Off	Off

#Known Limitations

Voice input requires an internet connection. Speech recognition happens on Apple's servers, so it won't work in airplane mode.
No wake word. You have to tap the mic to start — the app doesn't listen until you tell it to. Intentional.
Style accents are imperfect. The model takes a shot at what you ask for ("Irish accent", "Boston accent") but doesn't always land a convincing one. Accents are probabilistic; speed is exact.
Tapping speaker on an older message may regenerate audio if you changed your voice, style, or speed since it was first spoken. Expect a brief delay on first tap; subsequent taps of the same message use the cached recording.
Voice input transcription accuracy varies. Product names, brand names, and technical terms may not transcribe perfectly. Glance at the transcribed text before it sends if accuracy matters.

#Version History

Version	Date	What Changed
4	2026-06-07	Added Web note — voice is iOS-only; the web interface chat is text-only
3	2026-04-21	Added Speed control (seven-chip grid, deterministic rate via OpenAI TTS). Listening-ready chime now fires when the mic is actually hot, not at state transition. Three-dot throbber next to chat speaker icon during TTS generation. Mid-conversation voice/style/speed changes via natural language ("switch to Onyx", "speak at half speed") now apply instantly and are reflected in Voice Settings. Universal mute: tap any speaker icon to stop all audio. Per-message speaker icons added to the Voice Conversation overlay.
2	2026-04-19	Unified voice, voice style, and audio routing across chat speaker icon, auto-play, beacon narration, and the preview in Voice Settings. Voice and style changes now propagate instantly. Corrected voice list (six server voices; no more per-device voices or a speech-rate slider).
1	2026-03-01	Initial guide

Voice — Speaking & Listening