Privacy Tools Guide

Smart speakers are surveillance devices that happen to play music. They record audio constantly, analyze it for wake words, store transcripts indefinitely, and create detailed profiles of your behavior. The privacy risks extend beyond Amazon and Google to include device manufacturers, ISPs, and anyone on your network.

This guide explains what smart speakers actually collect, how to audit them, and what privacy-respecting alternatives exist.

What Smart Speakers Collect

Audio and Transcripts

Continuous recording:

Example capture:

User says: "Alexa, what's the weather?"
Captured audio: [5 seconds before "Alexa"] + "Alexa, what's the weather?"
Transcript stored: "what's the weather"
Audio file kept: Yes (encrypted, but Amazon controls decryption key)
Retention: Indefinite by default; you can delete manually but must opt-in to auto-delete

Wake word false positives:

Location Data

Smart speakers collect:

Why it matters:

Device Usage Patterns

What they track:

Aggregated profiles:

Shopping and Financial Data

Direct capture:

Inference:

Smart Home Device Network Topology

When you connect smart home devices (Philips Hue, Nest, August locks):

Network Traffic Analysis: What Leaves Your Home

Passive Monitoring Setup

To audit what your speaker sends, capture network traffic:

# On macOS/Linux with Wireshark installed
# Or use Charles Proxy for HTTP/HTTPS traffic

# Monitor all traffic from your smart speaker's IP
# 1. Find speaker IP: Router → Connected Devices → Find "Echo" or "Google Home"
# 2. Use Wireshark: Capture → Options → Interface selection
# 3. Filter by speaker IP: ip.addr == 192.168.1.100

# What you'll see:
# - Outbound connections to Amazon/Google servers
# - DNS queries (what domains are contacted)
# - Periodic "heartbeat" packets (always-on connection)
# - Data volume transferred

Common Outbound Connections

Amazon Alexa:

Primary servers:
- alexa-comms-service.amazon.com (command processing)
- metering.prod.us-east-1.amazonaws.com (usage metrics)
- cognito-idp.us-east-1.amazonaws.com (authentication)
- s3.amazonaws.com (downloading updates)

Frequency:
- Every 5 seconds: "heartbeat" keep-alive packet (confirms speaker is alive)
- Every 60 seconds: metrics upload (what you asked, when)
- Every 24 hours: device telemetry (usage patterns)

Data volume:
- Average: 150MB/month (mostly updates, some audio uploads)
- Spike hours: 2-4 AM (scheduled data sync)

Google Home:

Primary servers:
- googlemsyncd.clients.google.com (sync service)
- mcs.gstatic.com (message service)
- assistant.google.com (AI processing)

Frequency:
- Every 10 seconds: heartbeat
- On wake word: full upload of audio + metadata

Data volume:
- Average: 100MB/month (less than Alexa)
- Usage spikes: During music streaming (up to 500MB if playing music)

DNS Queries (Even More Revealing)

DNS queries show what you’re trying to access without needing to decrypt HTTPS:

Observed DNS queries from Echo device:
- bbc.co.uk (you asked about BBC news)
- weather.com (weather queries)
- aol.com (you own an AOL email)
- myfitnesspal.com (you linked fitness app)
- grubhub.com (you searched for food delivery)

Why this matters:

Mute: Hardware vs. Software

The Mute Button Problem

What people think:

What actually happens:

Muted: Microphone is disabled
  ✓ Local voice processing stops
  ✗ Uplink to Amazon still active (heartbeat)
  ✗ You can no longer give commands (device is non-functional)

Problem: You cannot tell if Amazon is still recording through software
  - Hardware mute only disables local mic
  - If there's a software vulnerability, Amazon could re-enable recording
  - No way to verify the mute works without technical analysis

Safer Approach: Unplug Entirely

Physically disconnected:
✓ No heartbeat packets sent
✓ No possibility of remote re-activation
✓ Complete privacy guarantee
✗ Device is non-functional (requires replug for voice commands)

Practical compromise:

Verify Mute Actually Works (Technical)

# While speaker is "muted," capture network traffic
# Command: tcpdump -i en0 -c 100 host 192.168.1.100

# If you see regular packets to Amazon, the mute may not be working
# If no packets appear, the network connection is actually disabled

# Safe assumption: Always assume the speaker is listening until physically disconnected

Privacy-Hardened Configurations

Amazon Alexa

If you insist on using Alexa, harden it:

1. Disable Voice Purchase

2. Auto-Delete Audio

3. Disable Skills You Don’t Use

4. Disable Drop In / Alexa Calling

5. Disable Smart Home Integration

6. Create Separate Amazon Account

7. Disable Device Analytics

Google Home

1. Turn Off Web & App Activity

2. Delete Search History Regularly

3. Disable Personalization

4. Restrict Smart Home Access

5. Use a VPN on Home WiFi

Privacy-Respecting Alternatives

Snips (Self-Hosted)

Setup:

# Requires Raspberry Pi (4GB RAM minimum) + microphone
# 1. Install Snips OS on Raspberry Pi
# 2. Configure offline voice models
# 3. Can control smart home devices locally

# No cloud = No privacy risk

Limitations:

Cost: ~$150 (Raspberry Pi + microphone + speaker)

Option 2: Open-Source Home Assistant

Home Assistant

Setup:

# 1. Install Home Assistant on Raspberry Pi or NUC
# 2. Add STT (speech-to-text) module: Wyoming Faster Whisper
# 3. Add TTS (text-to-speech): Piper TTS
# 4. Voice commands process locally, never leave your network

# Result: Full voice control, zero cloud tracking

Advantages:

Limitations:

Cost: ~$200-400 (hardware + no recurring fees)

Option 3: Minimal Use of Cloud Speaker

Compromise approach:

This works if:

Realistic privacy level: 60% reduction in tracking vs. default setup

Option 4: No Smart Speaker

Most private option:

When this works:

Household Strategy: Reducing Smart Speaker Risk

If multiple people in your household use smart speakers:

1. Segregate on separate WiFi network

2. Use WireGuard VPN for main devices

3. Disable microphones physically

4. Disable WiFi when away

What the law does NOT protect:

What you CAN do:

Reality: Large tech companies have legal teams that defend their practices. Individual complaints rarely result in action.

Practical Recommendation

If you own a smart speaker:

  1. Change the wake word to something less commonly spoken (reduces false positives)
  2. Auto-delete audio every month (Settings → Alexa Privacy)
  3. Disable skills and smart home integration you don’t actively use
  4. Unplug during private conversations
  5. Monitor network traffic monthly using Wireshark to verify what’s being sent

If you’re considering buying:

  1. Buy Home Assistant + Raspberry Pi setup (~$300) instead
  2. Or accept the privacy cost and use default hardened settings
  3. Or skip smart speakers entirely

If you’re privacy-conscious: Smart speakers are fundamentally incompatible with privacy. They exist to create data profiles about you. Any speaker with always-on microphone and internet connection is a potential surveillance device, regardless of manufacturer claims about local processing.

The most honest assessment: Smart speakers are conveniences you purchase by surrendering privacy. Some people accept this trade-off. Others don’t. That choice is valid either way, but it should be informed.


Built by Privacy Tools Guide — More at zovo.one