Hardware components | ||||||
 |
| × | 1 | |||
Software apps and online services | ||||||
| ||||||
Walkie-talkies never really went away. On a hiking trail with no cell coverage, at a maker event inside a noisy hall, or just playing around the neighborhood with kids, a push-to-talk radio is still the fastest way to say "where are you?" The problem is that proper handheld transceivers need a license or registration in many countries, and toy-grade ones sound terrible.
Meanwhile, every ESP32 board already contains a capable 2.4 GHz radio — one that is certified as a Wi-Fi device and therefore license-free to operate. So I asked a simple question: can a stock M5StickS3, with no extra RF hardware at all, become a real, usable voice transceiver?
The answer is ESP32Talkie: a pocket-size walkie-talkie that streams live voice over Espressif's ESP-NOW protocol in Long Range mode. No pairing, no router, no internet, no license. Turn on two units, press the button, and talk — up to about 1 km line-of-sight.
What it does- Real-time voice communication between two or more units over 2.4 GHz Wi-Fi frequencies
- 13 selectable channels (2412–2472 MHz, 5 MHz spacing) so multiple groups can operate side by side
- Push-to-talk with the front button, just like a classic transceiver
- Live status display: TX/RX state, channel, volume, RSSI, and a signal/power level bar
- Shake-to-adjust UI: flick the unit sideways to increase a value, vertically to decrease it — handy when your other hand is busy
- "Kero-Kero" (frog) voice effect modes that pitch-shift your voice up for fun (M1 normal / M2 double speed / M3 triple speed)
- Settings persistence: channel, volume, and voice mode survive power cycles
- Runs entirely on the M5StickS3's built-in microphone, speaker, display, IMU, and battery — zero external components
ESP-NOW is Espressif's connectionless 2.4 GHz protocol. Unlike normal Wi-Fi it needs no access point and no association handshake — a device can broadcast frames to any listener on the same channel, with latency in the low milliseconds. That is exactly the semantics of a walkie-talkie: whoever is on your channel hears you.
Two ESP-NOW features make this practical:
- Broadcast addressing. ESP32Talkie transmits to the broadcast MAC, so any number of receivers on the same channel hear the same audio. There is no pairing step at all.
- Long Range (LR) mode. The ESP32-S3 supports a proprietary PHY rate below 802.11b that trades bandwidth for sensitivity. With LR enabled, receive sensitivity reaches roughly −98 dBm, which is what stretches the range to about 1 km line-of-sight from a ~5 mW/MHz transmitter — all still within license-free Wi-Fi rules. (The M5StickS3's radio operates under Japanese technical conformity certification 219-259730; check your local regulations for equivalents.)
The trade-off of LR mode is throughput, which drives the audio design below.
Audio pipelineVoice is captured from the built-in PDM microphone at 16 kHz and transmitted as 8-bit samples, keeping the stream at 16 kB/s — comfortably inside the LR-mode budget while still sounding clearly intelligible for speech.
The pipeline looks like this:
[MIC] → I2S DMA → gain → 16→8-bit conversion → ESP-NOW packets (magic header "ESPT1")
↓ 2.4 GHz broadcast
[ESP-NOW RX] → packet filter → OutputBuffer (jitter buffer) → 8→16-bit → I2S → [SPEAKER]A few details that mattered in practice:
- Jitter buffering. ESP-NOW frames arrive in bursts and occasionally drop. Received PCM goes into a ring
OutputBufferthat absorbs timing jitter before the I2S speaker task drains it in fixed-size chunks (320 samples per wakeup — large enough to cut task-switching overhead, small enough to keep latency conversational). Version 1.4 focused on stabilizing this playback path and noticeably improved audio quality. - Packet filtering. Every payload starts with a magic header (
"ESPT1"). Since ESP-NOW broadcast will happily deliver frames from any nearby ESP-NOW device, the receiver drops anything that isn't an ESP32Talkie packet. This eliminated random noise bursts when other ESP-NOW gadgets were around. - Power saving. The Wi-Fi modem sleeps between activity (added in v1.3), which matters a lot on the M5StickS3's small LiPo battery. Peak consumption is about 1 W while transmitting.
The transport layer builds on the excellent transport and OutputBuffer classes from atomic14's esp32-walkie-talkie project, adapted and extended for ESP-NOW LR broadcast, channel switching, and packet filtering.
Because a walkie-talkie should also be fun: BtnB lets you cycle through three transmit voice modes. M1 is your normal voice; M2 and M3 resample the outgoing audio at 2× and 3× speed, producing a chirpy "frog voice" (kero-kero is the Japanese onomatopoeia for a frog's croak). Kids love it, and it doubles as an instantly recognizable "who's talking" marker between units.
User interface on a 0.85-inch screenFitting a full transceiver UI on the M5StickS3's tiny display took some iteration. The final layout has three zones:
- Top: Receive / Transmit status
- Middle: two-digit channel number, volume, and RSSI in rounded panels
- Bottom: a live level bar — incoming SIGNAL strength while receiving, mic POWER while transmitting
Controls follow a two-button-plus-motion scheme:
- BtnA — push to talk
- BtnB long-press — cycle the edit target: VOL → CH → MODE (the active target highlights in green, and deselects automatically after 5 s of inactivity)
- BtnB click, or a shake — change the selected value
The shake control uses the built-in IMU. The firmware reads the accelerometer, maps the axes to the current display orientation (so "sideways" is always sideways regardless of how you hold it), and applies a dominance test plus re-arm hysteresis and a 450 ms cooldown so that walking or PTT button presses never trigger false adjustments. A horizontal flick increases the value; a vertical flick decreases it. Adjusting the volume without looking away from what you're doing feels surprisingly natural.
All three settings — channel (1–13), volume (1–5), and voice mode — are written to NVS flash via Preferences, so the unit comes back exactly as you left it.
The project is a standard PlatformIO project (Arduino framework, espressif32, M5Unified):
src/main.cpp— UI: display layout, button/shake handling, settings persistencesrc/Application.cpp— the core: I2S mic capture, ESP-NOW TX/RX, jitter-buffered playback, pitch-effect processing, status displaylib/transport/— ESP-NOW transport (LR mode, channel control, magic-header filtering)lib/audio_output/— theOutputBufferjitter buffer
Two build environments are provided out of the box:
m5stack-sticks3(default) — M5StickS3, fully self-containedm5stack-atoms3-echo-base— M5AtomS3 + Atomic Echo Base speaker unit
Board-specific differences (mic gain, speaker driver, volume tables, display layout, which shake axes are active) are cleanly separated behind TARGET_* build flags in config.h, so porting to another M5Stack controller mostly means adding one more environment.
You need two of either supported device and nothing else.
- Clone the repo:
git clone https://github.com/pokibon3/ESPTalkie.git- Open it in VS Code with the PlatformIO extension.
- Pick your environment in
platformio.ini(m5stack-sticks3is the default; switch thedefault_envscomment for AtomS3 + Echo Base). - Build and upload to both units.
- Power both on, make sure they show the same channel, press BtnA, and talk.
That's the whole setup — no configuration files, no MAC addresses to exchange, no app.
Specifications- Frequency: 2412–2472 MHz, 13 channels at 5 MHz spacing
- Protocol: ESP-NOW Long Range mode (broadcast)
- Emission: G1D, D1D
- TX power: approx. 5 mW/MHz
- RX sensitivity: approx. −98 dBm
- Range: up to ~1 km line-of-sight
- Audio: 8-bit / 16 kHz sampling, ~1.0 W speaker output
- Controller: M5StickS3 (or M5AtomS3 + Atomic Echo Base)
- Power: 3.7 V LiPo, ~1 W max during TX
- Licensing: License-free (operates as certified Wi-Fi equipment)
- v1.0 — initial release
- v1.1 — M5AtomS3 + Atomic Echo Base support, Kero-Kero voice
- v1.2 — M1/M2/M3 voice modes, shake-gesture control
- v1.3 — Wi-Fi sleep for lower power consumption
- v1.4 — audio quality improvements (stabilized RX playback), packet filtering
- Audio codec upgrade (e.g. μ-law or ADPCM) for better quality in the same bandwidth
- More M5Stack targets — the config layer is already structured for it
- Group features such as a channel-busy indicator and unit IDs
ESP32Talkie started as a "can it even work?" experiment and ended up as a device my family actually uses. The M5StickS3 turned out to be a remarkable platform for this: microphone, speaker, display, IMU, button, battery, and a long-range-capable radio in one tiny certified package. Everything in this project — the UI, the DSP, the transport — is just software on top of hardware you can buy off the shelf.
The full source is MIT-licensed at https://github.com/pokibon3/ESPTalkie. Build two, pick a channel, and go talk to someone — no license required.
AcknowledgmentsThe ESP-NOW transport and output-buffer design builds on atomic14's esp32-walkie-talkie project — thank you for the great foundation.
Comments