How to Turn On Muse Voice Mode: The Step-by-Step Meta Didn't Write, What Realtime Voice and the Avatars Actually Are, Who Gets It Today, the Settings to Check, and What It Is Not

Everything here is from Meta's own pages read on November 6, 2026: the announcement, the help article that still describes the previous voice mode (and a different icon, which we explain), the data-controls and subscriptions help pages, the App Store release notes, the Realtime Avatar research post, and Alexandr Wang's launch post. Where the press adds something Meta didn't say, it's attributed. The screens in the diagrams are labeled reconstructions from those descriptions, not screenshots.

The 30-second version

• Turn it on: update the Muse app, open a chat, tap the phone icon top right, allow the microphone, talk. Say 'Goodbye' or 'Hang up', or tap the end button, to finish.
• What it is: Muse Realtime Voice, a live speech-to-speech conversation you can interrupt, with the agent doing your tasks in the background while you talk; plus Realtime Avatar, an animated face in sync with the voice, which you can show beside the chat, full screen, or not at all.
• Who gets it: US users of the Muse app on 'mobile, web, and Mac', rolling out from November 5. Free, Power and Maximum plans alike; Meta names no plan requirement and no per-minute cost. 'Coming soon' to Meta's AI glasses and the Muse Charm keychain. Not in Canada yet, not in the EU or UK (where Muse itself isn't), and Muse is 18+.
• Voices and faces: 'hundreds of voices'; change one any time from the avatar settings or by telling Muse in the chat what you want ('faster, slower, or even a specific accent'). The avatar can be a person-like character, an animal, or something you describe; Meta's demo is a dog named Jolly.
• Not this: the older Meta AI voice in WhatsApp and Instagram with celebrity voices, Muse phoning a business on your behalf, the 'Hey Muse' beta on glasses, or Meta's Voice Transcribe dictation model.
• Check two settings: Settings › Voice and Avatar (microphone and background listening) and Settings › Data controls › 'Help improve our AI models' (your voice is 'interactions' Meta can train on unless you turn it off).

What Muse voice mode is

Muse is Meta's personal AI agent, the one we rank first among personal agents. Until this week you typed to it. Voice mode makes it a call. Meta's words: 'Muse Realtime Voice technology lets you have fluid conversations in real time.' Responses 'arrive right away', you can interrupt mid-sentence, and the agent 'handles it in the background as you talk', which is the part that makes it different from a chatbot's voice toggle: you can say 'book the dentist for Tuesday and move my 3pm' and keep talking while it does both. Meta first showed it at Connect on September 23 with the line that it 'supports long conversations' and that you can shape the voice by describing it.

Realtime Avatar is the second half. Meta's research post gives the numbers: a 448×768 portrait video at 25 frames a second, generated live and synced to the voice, with about 870 milliseconds from the end of your sentence to the first frame of its reply; in blind tests raters preferred it to Runway's and HeyGen's live avatars roughly four to one and seven to one. In the app it shows up as 'expressive avatars' with three presence options, voice only, alongside the conversation, or full screen 'for face-to-face', with 'expressions and gestures' matched to what it's saying. The Verge's Jay Peters, who had it on his iPhone the day it launched: 'my Muse agent responds quickly over voice, though it occasionally pauses while it's looking up information or checking something for me'; it 'animates what it's saying well, though for my particular avatar, it's been a little uncanny' (he had set the avatar to look like Mark Zuckerberg, which he calls 'quite weird'). Wang's launch video shows his own Muse as a fluffy white dog; TechCrunch says users 'can also customize the avatar as they choose', person-like, animal or otherwise.

It does not watch you. TechCrunch's line, which Meta's post leaves out: Muse 'doesn't actually "see" the user, but it hears them and reacts based on their conversations.' The call is audio in, voice and face out; your camera stays off. Two things Meta did not say and we could not find anywhere: the voice latency itself (870 ms is the avatar figure) and which languages it speaks. Treat it as English until Meta documents otherwise.

Step by step on iPhone

• 1. Update the app. Open the App Store, search 'Muse from Meta', and update to version 14.3 or later. Its release note is the only one that mentions the feature: 'Try voice mode and expressive avatars, a new, natural way to interact with your Muse.' Versions 14.0 to 14.2, from November 1 to 4, were bug fixes. Requires iOS 18.
• 2. Open any chat with your Muse. New or existing. The composer at the bottom is for text; the call starts from the top.
• 3. Tap the phone icon in the top-right corner. Meta's instruction verbatim. If you see a headset icon there instead, you are on the older voice mode and the app hasn't updated; see troubleshooting.
• 4. Allow the microphone. The help page: 'You'll need to allow the Muse app access to your microphone to speak with your Muse.' iOS asks once; you can revisit it in iOS Settings › Muse or in the app's Settings › Voice and Avatar.
• 5. Talk. Speak as you would on a call. Interrupt when you want to. Give it tasks; it works on them while the call continues. You can leave the app: 'you can continue talking to your Muse in the background while you use other apps on your device', and iOS shows the orange microphone indicator while it listens.
• 6. Choose how the avatar appears. Voice only, beside the chat, or full screen. The control is in the call screen; Meta doesn't name the button, so look for the avatar itself or its presence option.
• 7. Pick a voice. Browse the options in the avatar settings, or just say what you want: Meta's examples are 'a New Yorker's energy' and 'a pirate's flair'; the App Store description says you can change it by describing it in chat. The Verge: 'by just asking your agent about what you want and letting it decide or by browsing through a bunch of options in the avatar settings.'
• 8. End the call. Say 'Goodbye' or 'Hang up', or tap the end button. Both are from Meta's help page.

Android, web and Mac: what's different

Android. Same steps. The Play listing for Muse from Meta (package name com.facebook.aura) was updated November 4 and Meta's post says 'mobile', but Google's page doesn't expose a version number, so if there is no phone icon, check for an update in the Play Store first. Meta's standing caveat applies: 'depending on your device's manufacturer your steps and what you see may differ.' Android's microphone-in-use indicator is the green dot.

Web. Sign in at muse.ai, open a chat, and look for the same phone icon. The one layout difference Meta documents is the menu: it is bottom-left on the web and top-left on mobile, which matters for the settings paths below. TestingCatalog saw the web voice build in September and described it as something that 'resembles a phone call with the agent'. Your browser will ask for microphone permission the first time.

Mac. The Mac app is a beta you download from Meta's site, not the Mac App Store, and it publishes no release notes; Meta's post names Mac as a voice platform and nothing else. If your copy has no phone icon, download the current build again. Two cautions from our earlier coverage: the Mac app had a real zero-day patched in September, and it is the app with computer-use permission, so a voice command there can drive your Mac. Keep 'ask before acting' on.

Pick a voice, pick a face, pick a mode

Meta's phrase is 'hundreds of voices to choose from', picked by mood and changeable any time. The avatar is part of what Meta calls your Muse's 'visual identity': its personality help page says you 'can ask your Muse in the conversation to change or update' it or 'manually edit this information'. In practice that gives you three handles: the avatar settings for browsing, a sentence in the chat for anything specific ('sound like a calm Scottish woman, a bit slower'), and the presence option during a call for how much of the face you want to see. One caution from Meta's own research page: the demo clips 'illustrate model capability and do not all reflect avatars available in the Muse app'.

Who can turn it on today

Glasses are the common confusion. There is already a 'Muse on your AI glasses (Beta)' that works today: in the Meta AI app go to Glasses › Device settings › AI assistant and select Muse, then say 'Hey Meta' or 'Hey Muse' (or a custom name under Manage 'Hey [name]', not on every model during the beta). That is a voice assistant, one request at a time, and in the beta it can't do music, calls, messages or navigation. The live, interruptible Realtime Voice call is what Meta says is 'coming soon to AI Glasses and Muse Charm'; the Charm keychain has no price and, per Android Authority, is expected to ship in December. Canada has had the Muse app since September 18 but voice is 'in the US to start' with no Canadian date; Wang announced the same evening that Muse itself had 'touched down in mexico', without saying whether voice came with it.

The settings to check before your first long call

Voice and Avatar is where the help page sends you for permissions: 'You control background listening' and 'You can review and change these permissions at any time in Settings › Voice and Avatar.' If you don't want the call to continue after you leave the app, this is the place.

Data controls › 'Help improve our AI models' is the one most people will want to look at now that they're talking instead of typing. Meta's data page says stored interactions 'can include text, voice, photos, things you create together' and that 'you decide whether we can use your interactions with Muse to train and improve our AI models'. Off stops the use and 'applies to previous interactions too'; on, Meta says it strips some personal details first. Nothing on Meta's pages describes a separate voice-transcript setting, a retention period for audio, or a way to delete voice alone, so the controls you have are the chat-level ones: delete a thread from the Chats tab, tap-and-hold a message, or Settings › Data Controls › Reset Muse, which 'cannot be undone'. Read Meta's own caveat before relying on deletion: 'After you delete something from Muse, Muse may still remember information it learned from what you deleted.' The App Store privacy label lists 'Audio Data' as user content linked to you, used for analytics, personalization and app functionality. Our safety guide has the fuller picture on what Muse keeps.

Usage is where plans live: Free with a weekly limit, Power at $20 for 500 million tokens a week, Maximum at $100 for 3 billion. Meta's subscriptions page doesn't mention voice at all, neither as included nor as metered. We'd assume a call draws from the same weekly pool until Meta says otherwise, and we'll update this when it does.

What voice mode is not

Troubleshooting

• No phone icon. Update the app (iPhone 14.3 or later); on Android and Mac, reinstall from the store or Meta's site. If it's still missing, you may be outside the US, the rollout may not have reached your account, or the account is under 18.
• A headset icon instead. That is the previous voice mode, which Meta's help page still documents ('tap the headset before speaking a prompt'). It works, but it is press-to-talk, not the live call. Updating replaces it.
• It can't hear you. Check the OS microphone permission for Muse (iOS Settings › Privacy › Microphone; Android app permissions), then the app's Settings › Voice and Avatar.
• It stopped when you switched apps. Background listening is a permission in Voice and Avatar; if it's off, the call pauses when you leave. When it's on, your phone shows the mic indicator the whole time, by design.
• It won't hang up. Say 'Goodbye' or 'Hang up', or tap the end button. Meta lists both.
• You hit a usage limit. Free has a weekly cap; a long call is still Muse usage. Settings › Usage shows where you are and the Upgrade button.
• You want it to stop learning from your voice. Settings › Data controls › turn off 'Help improve our AI models'. It applies retroactively per Meta.

What it looks like, and what early users say

Meta published a 58-second walkthrough the day of the launch, narrated by a Meta staffer named Ryan. It is the closest thing to an official tutorial that exists, and it shows the call screen and the avatar in motion:

Voice mode was in quiet early access for about two weeks before the announcement. The most useful independent review we found is from the YouTube channel Humanoid, published October 30 after 'about a week' with it, recorded as a live call with the creator's Muse, with chapters on the voices that weren't available yet, interrupting and being interrupted, and hallucinations on calls. Not sponsored, per the description; 18,772 views when we checked.

The official posts, verified on November 6:

Muse's post had 377,000 views and Wang's 249,000 by midday; Zuckerberg's Threads post, 'Now in the Muse app: talk with your Muse in real time. And pick from multiple voices,' had 600,000. The demand was visible in the App Store reviews the days before: a three-star review of version 14.2 on November 4 titled 'Needs voice' asked for 'a voice interaction so you don't have to type or use the keyboard. A way to activate it and be hands-free.' One thing to keep straight when you see the numbers: the most-viewed Muse clip of the autumn, Peter Yang's 1.5-million-view 'This phone call feature from @Muse is just ridiculous' from September, is Muse phoning a business for him, which is a different feature and the one in the table above.

What we could not verify

• An official written step-by-step. Meta's announcement gives one instruction, the phone icon; its help article, though updated on November 5, still describes the previous voice mode with a headset icon; the App Store note is one sentence; the walkthrough video has no captions. The steps above combine those with The Verge's hands-on and the on-screen words Meta's pages do use. We will replace them with Meta's own when it publishes one.
• The Android and Mac builds that carry voice. Google's listing shows no version string; the Mac beta has no release notes and Meta's download page returned an error when we checked.
• Latency, languages and cost. Meta states none of them for voice. The 870 ms figure is the avatar's; the 39 languages on the App Store are the app's interface, not the voice. That the camera stays off is TechCrunch's reporting, not Meta's text.
• Whether voice costs tokens, and how many per minute. Meta's subscriptions page is silent; we assume the weekly pool.
• Any voice-specific data retention or deletion control. None appears on Meta's Muse data page; the one-year voice-retention language you may have seen belongs to Quest and glasses notices, not Muse.