Guide 21 · Tutorial

Muse Voice Mode: Talking to Your Agent

Muse listens, talks back, and is gaining a face. How voice mode works today and what's coming next.

By Abhit Tomar · Updated November 21, 2026 · 17 min read

Jolly, Muse's cream avatar, with watercolor sound waves flowing across the scene
Illustration: museaicodes

Short answer

Muse supports voice conversations with real-time speech, plus newly added live video chat with customizable voices. Voice is becoming the primary way to direct your agent.

What does Muse voice mode do today?

The simplest way to use Muse may be to stop typing. Voice support, dictation, and voice notes are built into the app: tap the phone icon in the top-right corner and talk — Muse transcribes, understands, and replies out loud in real time while it keeps working on your task in the background. You can interrupt it any time, and there is no wake word and no hold-to-talk. It is live on mobile, web, and Mac for users in the US to start, after launching on November 5, 2026. It's the difference between operating software and briefing a colleague.

Voice shines for the fuzzy stuff — brainstorming, thinking through a decision, describing a complicated errand while your hands are busy. Save the keyboard for precision work like reviewing a contract or editing a draft line by line.

Muse voice mode process Four steps across two actors: you speak, Muse Voice Transcribe perceives the audio, Muse understands and acts as an agent, and you hear the spoken reply. 1 SPEAK 2 TRANSCRIBE 3 ACT 4 HEAR YOU MUSE YOU You speak tap phone icon to talk voice notes · dictation MUS Voice Transcribe audio perception model every 80 ms MUS Understands + acts transcribes · understands agent loop YOU You hear reply Muse responds out loud speaker STEPS 1 SPEAK 2 TRANSCRIBE 3 ACT 4 HEAR FLOW hand-off focal path
Diagram: museaicodes — how Muse voice mode works

Live video chat and custom voices

Voice just leveled up. Meta previewed live video chat and real-time voice conversations at Connect on September 23, 2026 — demonstrated publicly by Alexandr Wang on September 24 — and shipped them on November 5, 2026, as part of Muse Voice. Instead of a voice in a chat window, you talk to an animated presence — the Realtime Avatar system — with three display modes: voice only, the avatar alongside your conversation, or full screen.

The technical trick is that speech and video generate from a shared stream, so the avatar's voice, lip movement, and expressions stay synchronized. Strikingly, the system can animate a reference image during a live conversation — a photographic portrait, an illustration, an animal, or even an everyday object. Pair that with Jolly, and your agent is well on its way to having a face you chose. Voice is headed to AI Glasses and the Muse Charm next, but neither has it yet.

The tech underneath: Muse Voice Transcribe

Meta's voice push rests on Muse Voice Transcribe, a real-time audio perception model its Superintelligence Labs launched on September 1, 2026. It evaluates streaming audio every 80 milliseconds and handles speech-to-text, speaker identification, and endpoint detection in a single model — following conversations across more than 20 speakers and 25 validated languages, including mid-sentence language switching.

This is what powers dictation in the Muse apps today. The model was built for exactly the messy reality of talking to an agent: interruptions, accents, overlapping speech, hour-long sessions. When Muse seems to “just get” what you said, this is usually why.

Custom voices: voice choice is live, a personal library is not

Voice choice has now shipped. On November 5, 2026, Meta rolled out Muse Voice to all US users with hundreds of voice styles you can switch any time — or describe the voice you want in chat — alongside the realtime avatar. Picking a voice is live; what is still not here is a personal custom-voice library.

The full custom-voice flow Meta once previewed — generate your own voice and save it to a personal library — has not been announced as a shipped feature, and the earlier internal-testing reports have not turned into a release. Meta has also given no date for voice choice outside the US, or for custom voices on WhatsApp, smart glasses, or the Charm device. If a voice-library option appears, it will show up in the app's voice settings first.

Getting the most out of voice

  • Speak in outcomes, not menus: “book me a table for two Friday at 7 near downtown” beats “open OpenTable.”
  • Think out loud. Voice is ideal for rambling first drafts of a task — Muse is good at extracting the actual request.
  • Confirm before consequences. If a voice request could spend money or send messages, ask Muse to read back the plan first.
  • Use it hands-free moments: cooking, driving (parked or via car audio), walking — anywhere typing is awkward.
  • Switch to text for anything you need quoted precisely: addresses, numbers, names with unusual spellings.

Voice questions, answered

Does Muse have voice chat? Yes. Voice mode is built into the Muse app: tap the phone icon in the top-right corner and talk — Muse transcribes your speech and replies out loud in real time while it keeps working on your task in the background. Live video chat with the realtime avatar launched on November 5, 2026, for users in the US — see ‘Live video chat and custom voices’ above.

Does voice chat work on WhatsApp? No — voice mode is an app and web feature. On WhatsApp, Muse is text-only; see our WhatsApp guide.

When voice mode shines

Voice mode earns its keep in situations where typing is slow, awkward, or impossible. Brainstorming is the classic case: speaking your half-formed thoughts out loud keeps ideas flowing in a way that stopping to type each one never does, and Muse can reflect them back, ask follow-up questions, and help you shape the raw material. Hands-free scenarios are the other obvious win — cooking, driving, walking, or exercising — where a conversation you can carry on without looking at a screen turns dead time into productive time. It also works well for language practice, rehearsing a presentation, or thinking through a decision out loud with a patient sounding board. The common thread is that these are all conversational, exploratory tasks. Voice mode is at its best when you are thinking in real time and want a partner who keeps up with your pace.

How do I tune voice settings to my preference?

Before relying on voice mode, spend a few minutes in its settings. Most implementations let you choose a voice from a small set — pick the one you find easiest to listen to for long stretches, not just the one that sounds most impressive in a ten-second demo, because fatigue is real. Check the speech speed if the option exists; slightly faster than default often suits quick back-and-forth, while slower helps when Muse is explaining something complex. Language and accent settings matter too — if Muse consistently mishears certain words, verifying that the recognition language matches how you actually speak can fix a surprising number of errors. Also look for any auto-send or silence-detection options, which control whether Muse responds the moment you pause or waits for you to signal you are done. A longer pause threshold means fewer interruptions mid-thought; a shorter one means snappier exchanges. There is no correct setting, only the one that matches how you talk.

How do I get accurate transcriptions?

Voice mode lives or dies on transcription quality, and a few practical habits make a large difference. Speak at your normal pace — both rushing and over-enunciating tend to confuse recognition. Reduce background noise where you can: close a window near a busy street, step away from the kitchen fan, or move the phone's microphone closer to your mouth in a windy spot. Short, complete thoughts transcribe more reliably than long, rambling sentences with mid-stream corrections, so it helps to think in beats — say one idea, let Muse respond, then build on it. Names, technical terms, and uncommon words are the most likely casualties; if Muse mishears one, just correct it once plainly and move on rather than repeating it louder. And when accuracy really matters — an address, a number, a name you will act on — switch to text for that piece. Voice is for flow; text is for precision.

Ending sessions with a useful summary

One of the best habits in voice mode is to close the session deliberately. Before you hang up, ask Muse to summarize what you discussed: the decisions made, the ideas worth keeping, the action items. This takes thirty seconds and converts an ephemeral spoken conversation into something you can actually use later. Be specific about the format you want — a short paragraph recap, or a list of next steps if you are planning something. If you brainstormed, ask which ideas seemed strongest and why, while the reasoning is fresh. Then, when the session ends, save or copy the summary somewhere durable. Voice transcripts, where they exist, are searchable but often long and messy; your own end-of-session summary is the artifact you will actually return to. Treat the summary as the product of the session and the conversation as the raw material.

Limitations compared to text, and voice data privacy

Voice mode is not simply text with sound — it has real limitations worth knowing. You cannot easily scroll back and re-read an earlier answer, reference a specific paragraph, or copy a code snippet, so anything you will need to quote or paste later belongs in text. Long, detailed answers are harder to absorb by ear than on screen; if Muse starts a lengthy explanation, it is fair to ask for the short version now and read the full guide later. Interruptions and misheard words cost time, so high-stakes or detail-dense tasks — contracts, instructions you must follow exactly, anything with numbers — are better in writing. On privacy: voice mode processes your audio, which is inherently more personal than typed text. Check the app's privacy settings for voice-data controls — whether recordings are stored, whether they are used to improve the service, and how to delete them. As a sensible default, avoid dictating passwords, financial details, or other sensitive information by voice, especially on shared devices or in public where others can overhear. Use voice for the thinking, and keep the sensitive bits typed — or unsaid.