Amazon just made a decision for you. Starting with Alexa+, the "Do Not Send Voice Recordings" option has been removed from Echo devices. Your smart speaker now sends every voice command to Amazon's cloud — no exceptions, no opt-out. If that makes you uncomfortable, you are paying attention to something that matters.
Here is the thing: this was always the trade-off with cloud voice assistants, it just became more explicit. Every "Hey Google," every "Hey Siri," every "Alexa" — those wake-word detections, the commands that follow, the patterns of your daily life — are flowing to servers you do not own, analyzed by companies whose business model depends on knowing more about you. The question now is whether you are okay with that, or whether you want to build something different. Because you absolutely can build something different — and it works better than you might think.
This guide covers exactly how to reclaim your voice assistant privacy. Whether you want to tighten up your existing setup, switch ecosystems, or go fully local with zero cloud dependency, there is a practical path for you. We will look at what these systems actually collect, which privacy tier makes sense for your situation, and how to set up local voice processing that responds in under a third of a second with no data leaving your home.
Key Takeaways
- Amazon removed the voice recording opt-out from Alexa+ — all voice data now goes to the cloud with no exception
- Apple has the best privacy of the big three — Siri processes many requests on-device and HomeKit stays on your local network
- Home Assistant with local voice pipeline sends zero data to the cloud and responds faster than any cloud assistant
- 17,000+ people already run fully local voice using open-source tools Whisper.cpp and Piper on Home Assistant
- Three privacy tiers exist — lock down your current setup, switch to Apple, or go fully local — each is practical and worth doing
- Local processing is not a sacrifice — 0.3s response times and full smart home control with no internet dependency
What Voice Assistants Actually Collect
Let's be specific about what happens when you talk to your smart speaker. This is not speculation — it is documented in privacy policies and confirmed by independent security researchers. The picture varies significantly by platform.
Amazon Alexa — All Cloud, No Exceptions
Alexa records and stores all voice commands in Amazon's cloud. Amazon uses these recordings to train its speech recognition models, improve Alexa's responses, and build a profile of your habits, preferences, and home activity. Until recently, you could opt out of having your recordings used for model training, and you could disable sending recordings entirely. Both of those options are being removed with Alexa+.
The practical reality: your Echo device knows when you wake up (you say good morning), when you go to sleep (you turn off the lights), what music you like, when you are home, when you are not, what you watch on TV, what you ask about, and what products you consider buying. That is an extraordinarily detailed behavioral profile, and it lives on Amazon's servers indefinitely.
Accidental activations — where the device thinks it heard the wake word but did not — happen multiple times per day in most households. Those clips get recorded and sent to the cloud too, capturing conversations that were never meant for Alexa's ears.
Google Home — Cloud Processing, Ad-Linked Data
Google Nest devices use similar cloud processing. Your voice commands go to Google's servers, get transcribed, processed, and stored. Google does offer some on-device processing for the "Hey Google" wake-word detection phrase, which means the hot-phrase comparison happens locally. But the actual command — everything after the wake word — goes to the cloud.
Google's privacy controls are more developed than Amazon's: you can set voice data to auto-delete after three or eighteen months, and you can review and manually delete recordings through My Activity. But Google's core business is advertising, which means your smart home data feeds into the same system that serves you ads. The controls are limitations on a data collection system, not privacy-first design.
Third-party Google Assistant skills (called Actions) present an additional issue. When you enable a skill from a third-party developer, that developer can access your voice commands to their service. Not all third-party developers have robust privacy practices, and Google's oversight of what skills do with data is imperfect.
Apple Siri and HomeKit — Best of the Big Three
Apple processes many Siri requests directly on your device. With Apple Intelligence features on recent iPhones and HomePods, on-device language models handle a significant portion of requests without any data leaving your hardware. When requests do go to Apple's servers, Apple uses differential privacy techniques and does not link voice requests to your Apple ID.
HomeKit — Apple's smart home protocol — is fundamentally different from Alexa and Google Home. Your smart home automations, device states, and control commands stay on your local network. A HomePod or Apple TV serves as a local hub that processes commands without cloud dependency. You can control your lights, locks, and thermostats even when your internet is down.
The honest limitation: Apple's ecosystem is more expensive and more closed. You cannot use Siri with non-Apple devices the way you can use Google Assistant or Alexa. And "better than Amazon" is a low bar — Apple still collects data, it just treats privacy as a product feature rather than a cost center.
The Real Privacy Issues Nobody Talks About
Beyond the basic recording and storage issues, four deeper problems affect all cloud voice assistants.
Accidental activations are constant. A study by Northeastern University found that smart speakers accidentally activate 1.5 to 19 times per day depending on the device and background noise. Those clips are uploaded, transcribed, and in some cases reviewed by human contractors. Conversations you never meant to share are sitting in a database somewhere.
Third-party skills expand the data surface enormously. When you install a skill or action from a third party, that developer gets access to your voice commands directed at their service. The skill developer's privacy policy — not Amazon's or Google's — governs what happens to that data. Many skill developers have minimal privacy infrastructure.
Voice profiles are a surveillance layer. Amazon and Google build voice profiles that can distinguish between household members. This means the system does not just know what your household asked — it knows which specific person asked it. Over time, this becomes a detailed behavioral record for each person in your home, including children.
Data breaches are a permanent risk. All that voice data stored in the cloud is a target. Amazon, Google, and Apple have all experienced security incidents. Data that exists in the cloud can be subpoenaed, breached, or accessed in ways you never anticipated when you said "Hey Alexa, add milk to my shopping list."
The Local Processing Revolution
Here is what the mainstream tech press does not cover enough: local voice processing for smart homes is mature, practical, and already running in tens of thousands of homes. This is not a DIY experiment for hardcore hobbyists anymore. It is a real solution that works for normal people who want their home automation to stay in their home.
Home Assistant's local voice pipeline
Home Assistant — the open-source smart home platform — released a local voice pipeline that chains together several open-source tools to create a fully local voice assistant. Here is what runs entirely on your home hardware:
- Wake word detection: openWakeWord listens for custom wake words without sending any audio to the cloud. You can even train it on custom wake words.
- Speech-to-text: Whisper.cpp (a highly optimized version of OpenAI's open-source Whisper model) transcribes your voice commands locally. No audio leaves your network.
- Intent processing: Home Assistant's built-in conversation agent handles smart home commands locally. For more natural language, you can optionally add Ollama (a local LLM runner) that processes complex requests on-device.
- Text-to-speech: Piper converts responses back to speech locally, with dozens of voice options that sound surprisingly natural.
The result: you say "turn off the living room lights," the command is detected, transcribed, processed, and executed entirely within your home. Response time averages 0.3 seconds — faster than cloud assistants because there is no round trip to a server. No data leaves your network. No voice profiles are built. No accidental activation clips end up in Amazon's database.
Home Assistant Voice Preview Edition
For people who want dedicated voice hardware without building their own, Home Assistant released the Voice Preview Edition — a physical microphone device designed specifically to work with the local voice pipeline. It looks and works like a smart speaker, but every bit of processing happens locally. No cloud account required, no recordings stored externally, no data transmitted anywhere. It is exactly what Alexa should have been for privacy-conscious users.
The Voice Preview Edition works with the Home Assistant Green hub, which serves as the brain of your local smart home. The Green runs Home Assistant OS on dedicated hardware designed for always-on home automation — energy-efficient, quiet, and fast enough to run local speech recognition without any lag.
The Three Privacy Tiers
Not everyone needs to go fully local immediately. There is a realistic spectrum of privacy improvements you can make, starting with your current setup and going as far as you want. Here are the three tiers, in order of increasing privacy and increasing setup effort.
Keep Your Cloud Assistant, Lock It Down
This is the right starting point if you are not ready to change platforms. You can significantly reduce what cloud voice assistants collect without replacing your devices.
- Delete your voice history regularly — set auto-delete to 3 months in Google or Amazon settings
- Opt out of voice model training (while the option still exists on your device)
- Disable third-party skills you do not actively use — each one is an additional data pipeline
- Turn off voice purchasing or require a PIN
- Move Echo and Google devices out of bedrooms and private spaces — limit them to common areas
- Use physical microphone mute buttons when you do not need voice control
- Review and revoke permissions for any skill that has access to your calendar, contacts, or personal data
Privacy improvement: moderate. You are still sending data to the cloud — you are just limiting what gets stored and used. This is meaningful, but it is working around a system that was not built with your privacy as the priority.
Switch to the Apple Ecosystem
If you want substantially better privacy without leaving the mainstream consumer experience, Apple HomeKit with a HomePod Mini is the most practical upgrade. You get voice control, smart home automation, and a platform built on a business model that does not require monetizing your behavioral data.
- Apple HomePod Mini acts as your home hub — local processing, local automations, local device control Affiliate link — we earn a commission at no cost to you.
- HomeKit automations run entirely on your home network, no internet required
- Siri requests are anonymized and not linked to your Apple ID when sent to Apple servers
- Apple Intelligence on recent devices processes many Siri requests entirely on-device
- Matter and Thread support means your HomeKit devices can interoperate with other ecosystems
Privacy improvement: significant. HomeKit is architecturally different from Alexa and Google Home — local control is the default, not the exception. The trade-off is cost (Apple hardware is premium-priced) and ecosystem lock-in.
Go Fully Local with Home Assistant
This is the path for people who want zero cloud dependency. Home Assistant running on dedicated local hardware gives you complete smart home control with local voice processing, local automations, and no data leaving your home network.
- Home Assistant Green hub — plug-and-play hardware that runs Home Assistant OS, pre-configured and ready to go Affiliate link — we earn a commission at no cost to you.
- Local voice pipeline with Whisper.cpp speech recognition runs entirely on your hardware
- Supports 3,000+ device integrations including all major smart home brands
- Automations run even when internet is down
- 0.3s average response time — faster than cloud assistants
- Optional: add Nabu Casa cloud subscription ($6.50/month) for remote access while keeping local processing
Privacy improvement: complete. No cloud company holds your smart home data. The trade-off is initial setup time (30-60 minutes) and a slightly steeper learning curve compared to plug-and-play consumer devices.
How to Set Up Each Tier: Practical Steps
Here is what setup actually looks like for each tier, with specific actions and time estimates.
Tier 1 Setup — Lock Down Alexa or Google (30 minutes)
Start with Alexa if that is what you have. Open the Alexa app, go to More > Settings > Alexa Privacy. Here you can review and delete your voice history, manage skill permissions, and turn off features you do not use. Set voice history to auto-delete after three months — it will not prevent collection, but it limits how long Amazon holds your recordings.
Under Privacy > Manage Your Alexa Data, disable "Use Messages to Improve Transcriptions" and "Use Voice Recordings to Improve Amazon Services." These opt-outs may not survive the Alexa+ migration, but they are worth doing now while available.
For Google Home, visit myactivity.google.com, go to Data & Privacy > Web & App Activity, and set Voice & Audio Activity to auto-delete after three months. In the Google Home app, review each device's permissions and remove any skills or actions you do not actively use.
Tier 2 Setup — Switch to Apple HomePod Mini (1-2 hours)
Setting up Apple HomeKit starts with a home hub. The HomePod Mini is the most practical choice — compact, affordable for Apple hardware, and doubles as a decent speaker. Plug it in, open the Home app on your iPhone, and follow the setup flow. It handles home hub configuration automatically.
Next, add your devices. HomeKit supports Matter devices (which includes a growing number of switches, bulbs, sensors, and plugs), as well as devices from Apple HomeKit-certified brands. For devices that are not HomeKit-native, tools like Homebridge (a free open-source bridge) can bring many non-Apple devices into your HomeKit setup, though this adds some complexity.
Set up automations in the Home app — these run locally and will continue working even if your internet is down. Test this by turning off your router and confirming your automations still trigger. That is the moment you feel the difference between local and cloud.
Tier 3 Setup — Home Assistant with Local Voice (30-60 minutes)
This is easier than it sounds. The Home Assistant Green ships with Home Assistant OS pre-installed. Plug it into power and ethernet, go to homeassistant.local:8123 in your browser, and walk through the onboarding wizard. It automatically discovers many smart home devices on your network.
To add the local voice pipeline, go to Settings > Add-ons and install three add-ons: Wyoming Whisper (for speech recognition), Wyoming Piper (for text-to-speech), and openWakeWord (for wake-word detection). Each installs in about two minutes. Then go to Settings > Voice Assistants, create a new assistant, and connect these components. The entire voice setup takes about fifteen minutes after the base installation.
For hardware microphones, the Home Assistant Voice Preview Edition is the purpose-built option. You can also use any ESP32-based microphone running ESPHome firmware, which turns a $15 piece of hardware into a capable local voice device. The community has published dozens of pre-built firmware configurations you can flash without any coding.
Product Comparison: Which Voice Ecosystem Is Right for You?
Here is how the four main options compare across the dimensions that matter for privacy-conscious smart home users.
| Platform | Local Processing | Privacy | Ease of Use | Setup Cost |
|---|---|---|---|---|
| Amazon Alexa+ | None (removed) | Low | Very Easy | $50–100 |
| Google Home | Wake word only | Low–Moderate | Very Easy | $50–150 |
| Apple HomeKit | Most commands local | High | Easy | $100–200 |
| Home Assistant | 100% local option | Complete | Moderate | $100–300 |
Product Recommendations
These are the specific products we recommend for each privacy tier. Each has been selected based on privacy architecture, not just features.
Home Assistant Green Hub
Why it is the top pick: The Home Assistant Green is a purpose-built hub that runs Home Assistant OS out of the box. It is the foundation for a fully local smart home — local voice processing, local automations, and zero mandatory cloud dependency. It supports 3,000+ integrations including all major smart home brands, and the local voice pipeline runs on its hardware with sub-second response times.
This is the product that makes the privacy tier jump from Tier 2 to Tier 3 genuinely accessible to non-technical users. You plug it in, follow a browser-based setup, and you are running local smart home within thirty minutes.
Pros
- 100% local processing available
- No mandatory cloud account
- 3,000+ device integrations
- Local voice pipeline in minutes
- Works without internet connection
- One-time hardware purchase
Cons
- 30-60 min setup vs plug-and-play
- More options = more to learn
- Voice hardware sold separately
Apple HomePod Mini
Why it earns its spot: The HomePod Mini is the best privacy-respecting voice assistant in the mainstream consumer market. It runs HomeKit automations locally, Siri requests are anonymized when they do go to Apple's servers, and Apple Intelligence processes many commands entirely on-device. It is not cheap, but it is significantly better than the alternatives when privacy is the criteria.
It also works as a Thread border router, which is important if you are building a Matter-based smart home — Thread is the networking protocol that many Matter sensors and devices use, and you need a border router for them to work. HomePod Mini gives you that functionality alongside your voice assistant.
Pros
- Local HomeKit processing built in
- Siri requests anonymized by Apple
- Thread border router built in
- Plug and play setup
- Works without internet for local control
Cons
- Apple ecosystem required for full features
- Premium price point
- Limited non-Apple device support without Homebridge
Matter-Compatible Smart Home Hub
Why Matter matters for privacy: A Matter hub is the infrastructure layer that enables local device communication. When your smart bulb talks to your smart hub directly over your local network — instead of routing through cloud servers — that is Matter doing what it was designed to do. A Matter-compatible hub is the backbone of a privacy-respecting smart home regardless of which voice assistant you use on top.
Look for hubs with Thread support, as this is the low-power mesh networking protocol that Matter uses for battery-powered sensors and small devices. A smart home hub with Thread and Matter support gives you the flexibility to use devices from any brand while keeping communication local.
Pros
- Device communication stays local
- Cross-brand compatibility via Matter
- Thread support for low-power devices
- Works with Apple, Google, Amazon ecosystems
Cons
- Not all devices support Matter yet
- Some cloud features still need internet
The Privacy Tradeoff Is Smaller Than You Think
Here is the objection most people raise: "But cloud assistants are just better. More features, more integrations, easier to use." This was true five years ago. It is much less true today.
Home Assistant with local voice processing handles every common smart home command as well as Alexa — and handles them faster. "Turn off all the lights," "set the thermostat to 68," "is the garage door closed," "start the morning routine" — these all work perfectly locally. What you lose is things like asking Alexa to order toilet paper from Amazon or asking Google to check your Gmail calendar. If those features are central to your smart home use, the tradeoff is real. For most people, they are not.
Apple HomeKit gives you an even easier transition if you are already in the Apple ecosystem. The Home app works beautifully, Siri integration is seamless, and you get local processing without any technical setup beyond the initial device configuration. The limitations are ecosystem lock-in and premium hardware pricing — both real, neither insurmountable.
The other objection is reliability: "What if the local hub goes down?" Every system goes down sometimes. But here is the difference: when AWS has an outage, Alexa-controlled homes go dark because cloud processing is unavailable. When your local Home Assistant hub has an issue, you fix it yourself and get full control back — you are not waiting for Amazon to restore service. And because your automations run locally, many things continue working even if the Home Assistant UI is temporarily unavailable.
Privacy for Specific Situations
Families with young children
If children are in your home, voice assistant privacy is amplified. Those devices are capturing your kids' voices, their questions, their daily patterns. Amazon and Google build voice profiles that can distinguish between family members — including children. Apple explicitly does not build persistent voice profiles. Home Assistant local processing stores nothing externally. For families, the privacy tier jump from Tier 1 to Tier 2 or 3 is especially worth making.
Home offices and sensitive conversations
If you work from home and your calls or conversations touch on confidential information — client data, business strategy, personal finances — a cloud voice assistant in the room is a liability. The accidental activation rate (1.5 to 19 times per day per device) means sensitive conversations may be recorded and stored on third-party servers. The physical mute button is your friend here, or simply keep voice assistants out of your home office entirely.
People concerned about data requests
Smart home data has been subpoenaed in legal proceedings. Amazon has received law enforcement requests for Alexa recordings. Google has complied with geofence warrants that scooped up data from people in specific locations. If your voice assistant data is stored in the cloud, it is potentially accessible to law enforcement with an appropriate legal order. Local processing data is under your physical control — no subpoena to Amazon or Google can compel data that only exists on hardware in your home.
Making the Switch: A Realistic Timeline
You do not have to do everything at once. Here is a realistic migration path that spreads the effort over a few weeks without disrupting your daily life.
This weekend (1-2 hours): Do the Tier 1 lockdown on your current devices regardless of what else you plan to do. Delete voice history, set auto-delete, remove unused skills, move devices out of bedrooms. This is free, fast, and immediately meaningful.
Within two weeks: Order whatever hub you want to try — Home Assistant Green or HomePod Mini. Set it up in one room. Add a few devices. Get comfortable with the interface before migrating anything else.
Within a month: Move your most-used smart home devices to the new platform. Test your automations. Add the local voice pipeline if you went with Home Assistant. By this point you should have a clear sense of whether the new system works for your life.
Ongoing: As old Alexa or Google devices need replacing, replace them with Matter-compatible alternatives that work with your new local or Apple setup. You do not have to throw away working hardware — just stop adding to the cloud-dependent ecosystem and let it shrink naturally.
Ready to Build Your Private Smart Home?
Start with the hub that fits your privacy goal. Home Assistant Green for full local control — nothing leaves your home. HomePod Mini for the best mainstream privacy with Apple's local-first design.
Get Home Assistant GreenSee Apple HomePod Mini Affiliate links — we earn a commission at no extra cost to you.
Frequently Asked Questions
Technically, Alexa is always listening for its wake word — it has to be in order to detect it. Amazon says audio is only recorded and sent to the cloud after the wake word is detected, but independent researchers have found that accidental activations happen frequently, sometimes multiple times per day. With the removal of the "Do Not Send Voice Recordings" option from Alexa+, all voice data now goes to Amazon's servers with no opt-out. If you want genuine privacy, the answer is to either use a system with local voice processing or limit where Alexa devices are placed in your home.
Yes. Home Assistant running on local hardware (like the Home Assistant Green hub) gives you complete smart home control with zero cloud dependency. Your automations run even when the internet is down, your voice commands are processed locally using open-source tools like Whisper.cpp for speech recognition, and no data ever leaves your home. The Matter protocol also supports local device control across Apple, Google, and other ecosystems, though voice processing on those platforms still relies partially on cloud servers.
Home Assistant has become dramatically easier over the past two years. With dedicated hardware like the Home Assistant Green, setup takes about 20-30 minutes — plug it in, connect to your network, and walk through a browser-based setup wizard. Adding devices, automations, and integrations is largely point-and-click. The local voice pipeline (Whisper for speech recognition, Piper for text-to-speech) installs as add-ons with a few clicks. The biggest learning curve is that it gives you more control than cloud systems, which means more options to configure. Start simple and add complexity as you get comfortable.
Yes and no. Apple HomeKit processes most commands locally on your home network, and your automations will run without an internet connection. A HomePod or Apple TV acting as a home hub handles local processing. However, remote access to your home — controlling devices when you are away — requires internet. Siri voice commands use Apple's servers for language processing, though Apple anonymizes requests and does not build voice profiles the way Amazon does. For local-only control, you need to be on your home network or use Home Assistant.
Google Home collects and stores your voice commands, device usage patterns, and the times you control various devices. This data feeds into Google's advertising and personalization systems. Google offers some controls: you can auto-delete voice history after 3 or 18 months, and you can review and delete recordings manually in My Activity. However, unlike Apple, Google's business model depends on data, so privacy controls are limitations rather than the default. Google Nest devices support some on-device processing for the Hey Google detection phrase, but most requests go to the cloud.