Voice over ESP-NOW: building a routerless walkie-talkie pair with two ESP32s
Voice over ESP-NOW: building a routerless walkie-talkie pair with two ESP32s
Walkie-talkies are the oldest trick in the wireless handbook, so it says something when a new build of one gets attention. Aditya Verma's ESP32 walkie-talkie project does something quietly clever: it streams live voice directly between two ESP32 boards over ESP-NOW, Espressif's connectionless Wi-Fi framing protocol, with no access point, no router and no infrastructure of any kind — and it has been tested beyond 600 metres with an external antenna.
The build is complete in a way most hobby radio sketches are not: push-to-talk, twenty logical channels, an OLED interface, battery monitoring, signal indication, diagnostic logging, and automatic packet duplication to compensate for the fact that ESP-NOW provides no delivery acknowledgement. Full firmware, wiring and documentation are published in the ESP32-Walkie-Talkie repository on GitHub. Here is how the pipeline works, what the law says about using it in Australia, and when you would choose this over a proper LoRa mesh.
Why ESP-NOW instead of Wi-Fi
ESP-NOW is worth understanding on its own terms, because it is the answer to a question that comes up constantly in ESP32 projects: how do I get packets from board A to board B without a network? Espressif's own documentation describes it as a connectionless protocol where application data rides in vendor-specific Wi-Fi action frames, transmitted directly from one device to another, protected by CCMP encryption — the same cipher standard Wi-Fi uses. The practical consequences of "connectionless" are what shape this project:
- No association handshake, so a radio can transmit the instant it wakes. That is ideal for push-to-talk latency.
- No access point to range-check you, so two boards negotiate nothing; the limit is pure RF path loss, which is why the external U.FL antenna connector matters and why 600 metres line-of-sight is plausible.
- No acknowledgement at the application layer by default. The link layer reports success or failure per frame, but if a frame is lost, nothing retransmits it. Verma's firmware handles this by duplicating packets automatically when signal quality degrades — crude, effective, and honest about the medium.
For voice, the bandwidth arithmetic works out beautifully. ESP-NOW frames carry up to 250 bytes per packet in the classic version, and 16 kHz of mono audio compressed with IMA ADPCM — the lossy-but-simple codec the firmware uses — fits comfortably inside one frame per packet with room to spare. At the default 1 Mbps basic rate, a steady stream of voice packets is a rounding error on what the radio can push. The challenge is never bandwidth; it is timing, which is why the capture, compression and transmit path all live in firmware with tight loop discipline.
The hardware pipeline
Each handset is built around an ESP32-WROOM-32U module — the U suffix is the external-antenna variant, and that single letter does most of the range work. Around it:
- An INMP441 I2S digital microphone for capture. I2S microphones deliver digital audio straight off the membrane, which sidesteps the noise floor of the ESP32's analog inputs entirely — a far better starting point than an electret capsule into an ADC.
- A MAX98357A I2S class-D amplifier and speaker for playback, the standard pairing in this ecosystem because the amp takes the digital stream and needs no analog volume stage — a 10K potentiometer handles that.
- A 0.96-inch SSD1306 OLED for the menu system, channel display and signal indicator.
- Push-to-talk button, five menu buttons, battery voltage monitoring and a 3D-printed enclosure.
The author frames the device as something like a wireless-oriented Flipper Zero — a handheld platform that happens to be a walkie-talkie today and a remote controller, sensor node or robot link tomorrow, with three exposed UART pins for expansion. That reusability framing is sensible: a unit with a screen, buttons, battery and a radio is the expensive part of many projects, and firmware is the cheap part.
The legal picture in Australia
This is the part that matters for Australian builders, and the news is good: a peer-to-peer voice link over the 2.4 GHz Wi-Fi band using certified Espressif modules sits squarely inside the equipment class Australians use every day. ESP-NOW traffic is ordinary 802.11-band transmission from a compliant module, so it is treated the same way as any Wi-Fi device under the ACMA's framework for low-interference 2.4 GHz operation. The cautions that do apply are the ordinary ones: do not attach external amplifiers that push you beyond the power limits for the band, keep to the 2.4 GHz channels the module's firmware uses, and do not attempt to bolt on longer-range spectrum without proper certification. Voice traffic is also worth thinking about from the other direction — it is broadcast in the clear unless you enable the protocol's encryption keys, so treat a two-board link as a shout across the park, not a private line. Pairing ESP-NOW's CCMP support with per-unit master keys is straightforward and closes most of that gap.
The not-legal-advice caveat applies: if you plan to push range with non-standard antennas or higher gain, check the current ACMA rules for the band first.
When ESP-NOW, when LoRa
The honest comparison for Australian readers is with the LoRa mesh we cover constantly, because both are infrastructure-free links and the choice has real consequences. ESP-NOW gives you real-time audio — something LoRa cannot do, since its data rates are far too low for voice streaming. It gives you hundreds of metres cheaply. What it cannot give you is multi-hop range: an ESP-NOW packet does not route through intermediate nodes (nothing stops a relay firmware from forwarding, but you are building that yourself), and at 2.4 GHz, foliage and buildings eat the path fast. A Meshtastic starter kit on 915 MHz LoRa will tunnel text messages through hills, across kilometres, through nodes that relay for each other — at the cost of tiny bandwidth and no voice. Verma's own project page markets the same hardware as a Wi-Fi/IoT handheld controller platform, which is the other honest use: short-range, high-rate, direct links between devices you control.
For the parts-bin builder, the controller side of this build is entirely covered by a standard ESP32 dev board plus the two I2S modules — the INMP441 microphone and MAX98357A amplifier are both commodity parts with abundant documentation. Our walk-through of off-grid mesh fundamentals covers the LoRa side if your use case drifts past a few hundred metres or needs to survive a hill.
Takeaways
Three lessons from this build generalise beyond walkie-talkies:
- Connectionless does not mean unreliable — it means you own reliability. Packet duplication on weak signal is a legitimate engineering answer, and the diagnostic logging the firmware includes is how you find out whether it is working.
- I2S audio is the quality unlock on ESP32. The INMP441/MAX98357A pairing turns a microcontroller into a competent audio device with no analog front-end design at all. It is the right default for any ESP32 audio project, from intercoms to bird-call detectors.
- The 2.4 GHz band is a free underused resource for device-to-device links. While everyone routes their ESP32s through a home router, direct ESP-NOW links offer low latency, long range with a proper antenna, and zero infrastructure — legally, with the same certification as any Wi-Fi gadget.
Two boards, four modules, an enclosure and an afternoon: that is the whole barrier between you and a working radio pair that answers to nobody's network. In a world of app-registered, cloud-relayed communication gadgets, that is worth more than the modest parts bill suggests.