M5Stack Atom VoiceS3R Smart Speaker Dev Kit

M5Stack's Atom line has always charmed me with form factor, but voice AI on a sub-$20 module the size of a sugar cube? That seemed ambitious. After two days of using the Atom Echo S3R as a real voice device — connecting it to OpenAI's Realtime API, streaming bidirectional audio at 24 kHz over Wi-Fi, driving a NeoPixel ring, all while running an animated state machine for a 4-year-old. This thing punches absurdly above its weight. What's actually inside The S3R packs an ESP32-S3-PICO-1 with 8 MB octal PSRAM and 8 MB flash into a 24×24×16 mm cube. That's the dual-core 240 MHz Xtensa LX7 you want, paired with the kind of RAM headroom that lets you do things like buffer 40 seconds of audio in a queue without thinking about it. The audio path is properly engineered: an ES8311 mono codec with both ADC and DAC on a real I2S bus, plus an NS4150 Class-D amp with separate enable control so you can power the speaker down cleanly. There's a usable mic, a surprisingly loud little speaker, native USB Serial/JTAG (no flaky USB-to-UART chip), a Grove port, side breakouts, an IR transmitter, and a programmable button.