Straight Talk: Everything You Need to Know About Sonos 27voice 🎙️
Howdy, everyone 🤠
Back in the Beam Ultra deep dive, I promised a closer look at Sonos 27voice, and I'm glad that the time has come. Sonos 27voice is our new, optional AI voice assistant for music and system control. It launches in Early Access later this Fall in US English, for eligible households in the United States at no extra cost, as an opt-in upgrade to Sonos Voice Control.
A quick naming note before we dive in: as we introduce Sonos 27voice, you’ll also start seeing what you know today as “Sonos Voice Control” referred to as “Classic Sonos Voice Control". Nothing about how it works is changing; the new name simply helps distinguish it from Sonos 27voice.
The short version of what changed: we took what we learned from Sonos Voice Control and built a new voice assistant using Sonos technology together with select third-party AI models, giving it the ability to understand more natural, conversational requests and to answer general knowledge questions. Let's get into the technical details, what it can do, and what you should expect from this early build.
Two voices, one system
Before we get to the fun stuff, let's address the obvious question: what happens to the Sonos Voice Control you already use?
We're keeping both. Sonos Voice Control isn't going anywhere. It will continue to process your voice commands locally on the speaker, which makes it incredibly fast and private. But local processing has a low ceiling on how complex a command can be. To understand natural language, the assistant needs the power of the cloud.
Sonos Voice Control stays local, and Sonos 27voice goes to the cloud. You can swap back and forth as you please.
Sonos Voice Control vs. Sonos 27voice
The question I expect to see most: what actually changes if I upgrade? Here’s the breakdown:
| Feature | Sonos Voice Control | Sonos 27voice |
|---|---|---|
| Price | Free | Free |
| Wake word | "Hey Sonos" | "Hey Sonos" |
| Follow-ups without the wake word (following certain commands) | Yes, limited | Yes, full featured |
| Group, ungroup, move music | Yes | Yes |
| Queue manipulation | None | Yes: Add to queue / play next & Queue generation for highly specific requests |
| Hue and Lutron | Yes | Yes |
| Music services | Same list | Same list |
| Phrasing | Specific commands | Natural, vague, or chained requests |
| Dialogue | No | Yes, ask a follow-up or correct it |
| Context-aware | No | Yes, uses recent conversational context for a short period of time to support subsequent requests |
| Open-ended/general knowledge questions | No | Yes |
| Where requests are processed | On the speaker | Sonos cloud infrastructure |
| Languages | English (US), French | English (US) |
| Weather | Your set location only | Any location |
| Portables off Wi-Fi (Bluetooth mode) | Basic requests | Not supported |
A Good Understanding
With first-generation voice assistants, we were trained to speak in the assistant’s preferred structure: "Play (album) by (artist) on (service) in (room)." It’s logical, but it’s not how we normally communicate. It takes training, and it’s easy to make mistakes.
With Sonos 27voice, you say what you want the way you would describe it to a friend. No need to speak in rigid commands anymore, no talking in circles…just straight talk. For music fiends like me, it feels like talking to that friend that seems to know everything about every genre. But it goes beyond that, with general-purpose “ask me anything”-style questions available for the first time.
- Run the whole house in one sentence: Rooms, saved groups, and zones work the way they do in the app, so you can say "Play lo-fi electronic in the kitchen and something the kids will like upstairs," or "Play jazz everywhere except the TV room." One request, two rooms, no follow-up needed.
- Ask for music the way you remember it: You don't need exact track titles. You can ask for "the album with the baby swimming on the cover" or "that song Dolly wrote about her coat."
- Conversational follow-ups: If you ask a question and the answer isn't quite right, you don't have to start over. Just say, "No, the acoustic version," or "Actually, turn it up a bit more."
- Even when you get it wrong, Sonos 27voice gets it right: Tell it to "play that Pearl Jam song that goes 'Here we are now, entertain us.'" It will correctly play that little-known Nirvana track.
- Ask about science, history, trivia, pop culture, and the world at large: I’ve asked everything from “who won the Super Bowl this year?” to “tell me some interesting facts about quantum mechanics” and “what’s in the news today?”
It's built to understand the way we actually speak, with an “um” or “er”, or an “oh wait, I mean…” to adjust mid-sentence. And now that it can go beyond playback controls to answering questions and providing information, I find that 27voice can act as my primary voice assistant around the home.
To get an idea of what went on behind the scenes to make this all possible, I spoke with Nathan Hart, Principal Product Manager for Sonos 27voice. He had this to share:
“A big part of what makes it so good at understanding music and Sonos systems is the work we did to fine tune our in-house Large Language Model (LLM). We created our own data set: thousands of scenarios that help our LLM understand the right thing to do in a given situation with a given query. These were all written by hand by our machine learning engineers and linguists, and are a big part of the “secret sauce“ in Sonos 27voice.”
- Nathan Hart, Principal Product Manager for Sonos 27voice
Under the Hood
Sonos 27voice is a new voice assistant from the ground up. It uses a large language model (LLM) fine-tuned by Sonos to excel at understanding music and Sonos systems, and taking the right action for the context and the query. That LLM runs in the cloud and interfaces with other services - a new content resolution service, cutting edge text-to-speech for vocal responses, etc. to deliver the full experience.
With Sonos 27voice, wake-word detection still happens on the speaker, and your voice isn't sent to the cloud until it hears “Hey Sonos.” After that, your request goes to our cloud infrastructure for speech recognition and natural-language processing, where Sonos technology processes it and determines what you meant and how to respond. For certain capabilities, including some open-ended requests, we also use third-party AI services.
It’s still built with privacy at the heart of it. We don't save your voice recordings or transcripts; they're used only to process your request and are deleted once it's fulfilled. I don’t normally recommend reading a Privacy Policy for fun, but I took some time to read ours recently, and it breaks down the process in clear, technical descriptions. Head over and check it out if you’re interested in that sorta thing.
With all of that being said, one thing from the fine print: it's a home audio assistant, not a vault. Don't use it to process or retain your account numbers, passwords, medical data or other sensitive information.
More information
- Sonos 27voice will be released in Early Access. It's opt-in, it will change, and it will occasionally miss. Your feedback is a big part of why it launches this way.
- Early Access is only available to households in the US. Keep in mind, not everyone will be selected. If you are selected, there will be a delay between signing up and gaining access.
- English only (for now). Sonos Voice Control also speaks French, and since the upgrade applies to the whole system, a French-speaking household should stay on Sonos Voice Control until more languages arrive.
- It needs the internet. Because Sonos 27voice relies on cloud processing, it requires an active Wi-Fi connection. If you take a Move 2 out to a campsite, Sonos 27voice will not work.
- There’s a new voice. We’ve upgraded our text-to-speech tech and have a new voice for Sonos. Sonos Voice Control will keep its original voice.
- One Sonos Voice per system. You can't run Sonos Voice Control in one room and Sonos 27voice in another. Upgrading and downgrading are always available.
- Google Assistant can't share a speaker with it. This is the same as Sonos Voice Control. Sonos 27voice will happily co-exist with Amazon Alexa on the same speaker. Google Assistant can still live on a different speaker in the same system.
- The queue is add/play-only. It'll play, pause, resume, skip, or even add to your queue, but it won't remove, reorder, or save it as a playlist.
- It's a music-focused assistant. No alarms (timers, yes), no announcements or intercom, no routines, no calls, no shopping. And on trivia, treat it like a well-read friend rather than an encyclopedia.
- Sonos Voice Control will continue to be available. Those of you who participate in Early Access will notice that its name in the app will change to Classic Sonos Voice Control - it’s the same assistant, just with a new name to help disambiguate the two.
Wrap-Up
Sonos 27voice lands in Early Access in the coming weeks, US-English first, included at no extra cost on supported Sonos products with a microphone. We’ll make sure to post a reminder when you can sign up. When it opens, the upgrade lives in the Sonos app via Settings → General Settings → Voice Assistants, and the downgrade option sits in the same place if you want to swap back and forth.
It's also one of three ways you can use AI with Sonos 27: Sonos 27mcp (in Early Access now) lets the chatbot you already use control your system, and Sonos Custom Agents (coming in 2027) will let you build your own.
I'll be in the comments to answer what I can and chase down what I can't. Tell me how you end up using it, and don't hold back on the honest feedback; that's what Early Access is for.
More where that came from 🤠