Best Home Assistant Voice Satellites 2026: ESP32 & Local

The Raspberry Pi 5 8GB folds HA server and Wyoming satellite duty into one board. The Cortex-A76 quad-core at 2.4 GHz with 8 GB LPDDR4X memory runs Home Assistant OS, Whisper speech-to-text, Piper text-to-speech, and the Wyoming satellite add-on at once. Home Assistant's official voice documentation lists a Raspberry Pi 5 with a USB microphone as one of the supported Wyoming satellite setups, establishing it as a fully self-hosted alternative to Alexa and Google Assistant when wake-word, Whisper, and Piper all run on-board. General-purpose desks like CNET and Tom's Guide do not benchmark a Pi-as-voice-satellite build, so the supported-hardware list comes from HA's own docs. The SHE Privacy Score is 9.8: Local Control 10.0 because all transcription and synthesis stay on-board; Data Policy 10.0 reflecting the Raspberry Pi Foundation's charitable structure; Encryption 9.0 via HTTPS with user-configured certificate authority; Cloud Independence 10.0. Budget a 5A USB-C supply and roughly 30 min to flash HA OS to a microSD. Architectural constraint: the Pi 5 alone has no audio capture. Pair it with the Seeed Studio ReSpeaker USB Mic Array for XMOS-class far-field acquisition, or deploy as a compute node while Pi Zero 2 W satellites cover distant rooms.

The ESP32-S3-BOX-3 is the dedicated puck Home Assistant's official voice documentation points buyers toward. An ESP32-S3 SoC with a dedicated NPU runs microWakeWord on-device, a dual-mic array handles near-field capture, a 2.4-inch LCD touchscreen shows mute state, and a built-in speaker handles TTS playback — flashing the official HA voice firmware in about 5 min over browser-based improv-serial and claiming in HA in about 15 mins. XDA Developers' 2026 local-voice-assistant coverage frames the ESP32-S3 as a leading ready-to-use replacement for cloud-dependent Echo and Nest pucks, and HowToGeek's Meet Nabu series charts the open-source Alexa alternative's local-voice shift. Coverage like that lives on enthusiast desks — Engadget and PCMag review cloud Echo and Nest pucks, not this one. The SHE Privacy Score is 9.7: perfect Local Control (10.0) because the NPU runs wake-word on-chip before any audio leaves the device, perfect Data Policy (10.0) because Espressif is a chip vendor with no cloud service, Encryption 8.5 from ESPHome config, and Cloud Independence 10.0. The dual-mic array is good at near-field range but, lacking an XMOS DSP, is weaker than four-mic arrays for far-field capture in large rooms. For far-field noise cancellation, pair a Pi with the Seeed Studio ReSpeaker USB Mic Array instead.

The Seeed Studio ReSpeaker USB Mic Array is the closest Amazon-available analogue to the HA Voice Preview Edition's mic performance. No major outlet — CNET, Tom's Guide, or PCMag — has published a SKU-level review of this mic array as a Wyoming satellite, so the supported-hardware signal comes from Home Assistant's own docs rather than a mainstream desk. Four microphones, an XMOS XVF3000 DSP, hardware acoustic echo cancellation, and 12 programmable LEDs deliver the best-mic option in this roundup for buyers who already have a Pi or mini PC as the HA server. HA's documentation lists it as a supported Wyoming satellite microphone, and the Home Assistant Community reports that "the XMOS chip makes a night-and-day difference versus a raw USB mic — far-field works from across the room," enabling reliable barge-in while a speaker plays music. The SHE Privacy Score weighted composite is 9.6: perfect Local Control (10.0) because the USB host runs wake-word locally, perfect Data Policy (10.0) because Seeed Studio is a hardware-only vendor, Encryption at 8.0 for firmware-driven USB transport, and perfect Cloud Independence (10.0). It is not self-contained — it needs a USB host, a speaker, and a Wyoming add-on configured on that host.

The Seeed Studio ReSpeaker 2-Mics Pi HAT converts any 40-pin Raspberry Pi — including the Pi Zero 2 W, Pi 4, and Pi 5 — into a voice satellite with the only wireable GPIO mute / push-to-talk button of any Amazon-available pick here. It integrates a WM8960 stereo audio codec, dual microphones for near-to-mid-field acquisition, a 3.5 mm headphone output, a JST 2.0 speaker terminal, addressable RGB status LEDs, and a GPIO button wired for mute or push-to-talk. Home Assistant's documentation calls it "a common starting point for DIY voice satellites on Raspberry Pi," and Home Assistant Community hardware threads consistently identify a Pi Zero 2 W paired with this ~$40 HAT as the lowest-cost approach to a capable room satellite. The SHE Privacy Score weighted composite is 9.6: Local Control is 10.0; Data Policy is 10.0 from established hardware-vendor lineage; Encryption is 8.0; Cloud Independence is 10.0. Trade-off: with two microphones and no XMOS DSP, it relies on software echo cancellation and drops commands at far-field distances in open-plan rooms. Budget roughly 30 min to image the microSD and assemble the HAT, which runs on the host Pi's 2.5A supply. Pair it with the CanaKit Raspberry Pi Zero 2 W Starter MAX Kit for a finished satellite.

The CanaKit Raspberry Pi Zero 2 W Starter MAX Kit bundles the Pi Zero 2 W alongside a 64 GB microSD, a power supply, and a case — everything necessary to flash a Wyoming satellite image and direct it toward your Home Assistant server. A quad-core Cortex-A53 at 1 GHz and 512 MB RAM constrain it to satellite-only operation, yet sub-2 W idle consumption means a ten-room deployment accumulates negligible expense compared to always-on commercial speakers. Home Assistant's documentation states "Wyoming Satellite software runs well on the Raspberry Pi Zero 2 W," and the Home Assistant Community calls "Pi Zero 2 W plus ReSpeaker 2-Mic HAT" "the canonical 'multiple rooms' voice satellite build." No mainstream desk like TechRadar or ZDNet benchmarks a multi-room Pi array. The weighted SHE Privacy Score composite is 9.6: Local Control 9.5 because the Zero 2 W streams audio to the server; Data Policy 10.0 from Raspberry Pi Foundation lineage; Encryption 9.0 via Wyoming over LAN with HTTPS; Cloud Independence 10.0. One caveat: only 2.4 GHz Wi-Fi 4, which can encounter contention on congested networks; budget 30 min to image the microSD and a 2.5A supply per node.

The M5Stack Echo Pyramid Smart Speaker Base Atom Series accepts an M5Stack Atom, AtomS3, or AtomS3R ESP32 controller to become a hardware echo-cancelling voice satellite. Unlike software-only satellites, the base handles echo cancellation and noise suppression in hardware; M5Stack's product page specs the base with on-board echo cancellation and noise suppression plus two RGB touch sliders, in one of the most compact Atom-based enclosures available, and the Home Assistant Community notes that "once you have an Atom S3, it's a drop-in base — half the size of an ESP32-S3-BOX." Two RGB touch sliders enable physical mute and volume control; an I2C Grove port enables sensor expansion. The SHE Privacy Score weighted composite is 9.5: Local Control 9.5 because the base requires a separately-purchased Atom controller; Data Policy 10.0 because M5Stack is a hardware-only vendor; Encryption 8.5 via ESPHome configuration; and Cloud Independence 10.0. Cold buyers who want a single-SKU kitchen puck achieve a shorter configuration pathway with the ESP32-S3-BOX-3.

The Anker PowerConf S3 Speakerphone is the only non-dev-board pick — a commercial USB-C conference speakerphone repurposed as an HA Assist satellite. Six mics, full-duplex audio, a battery rated for 24 hours of call time, and a professional look make it the one pick here suited to a desk versus a dev bench. Anker's product documentation states that "six microphones capture every voice clearly, even in enhanced mode for crowded spaces," and the Home Assistant Community calls it "the stealth pick — a desktop speakerphone that doubles as an HA satellite." The SHE Privacy Score composite is 8.8 — the lowest here — because Anker ships a PowerConf+ companion app that collects telemetry when installed. Used purely as a USB audio-class device with HA, it generates no telemetry; Data Policy scores 7.5 versus 10.0 for pure hardware vendors, while Local Control is 9.5, Encryption 8.5, and Cloud Independence 10.0. As a plug-and-play USB audio-class device it claims into HA in under 10 min, and the six-microphone array sustains full-duplex capture well beyond the reach of any dual-mic pick here, enabling dependable barge-in throughout TTS playback. No hardware mute switch is fitted; the illuminated ring shows listening state visually.