Layer: L7-Application · Board: Seeed XIAO ESP32-S3 Sense · Core: arduino-esp32 3.x · PSRAM required

Turn the board into a tiny webcam-with-a-mic: browse to it and you get a live MJPEG video from the OV2640 camera and a live audio player fed by the onboard PDM microphone - both streamed by the server.

What this example teaches

Every response here goes out through one library call:

server.send_chunked(slot, 200, content_type, source, ctx);

send_chunked pages an unbounded body onto the wire in constant memory - it pulls a few KB from your source callback each time the TCP send window has room, and resumes on the next loop as it drains. So the body can be far larger than any buffer, and it is never truncated at the window. The two sources here are live hardware producers:

Video (/video): each call frames one JPEG into a multipart/x-mixed-replace MJPEG stream - the exact format a browser <img> plays as motion video. The source is a small state machine: emit the part boundary + headers, then the JPEG bytes, then grab the next frame. It never returns 0, so the stream runs until the browser disconnects.
Audio (/audio.wav): emit a 44-byte WAV header once, then hand out continuous 16 kHz / 16-bit PCM read straight from the microphone. It streams at the microphone's real-time rate (~32 KB/s), which is exactly right for audio.

The library owns the HTTP framing and flow control; you only supply the bytes. The same send_chunked path also serves large files, generated reports, and firmware images.

Routes

Route	What it is
`/`	a page showing the live video and an audio player
`/video`	MJPEG video stream (`multipart/x-mixed-replace`)
`/photo.jpg`	a single still JPEG
`/audio.wav`	continuous live audio (a never-ending WAV)
`/clip.wav`	a fixed 2-second WAV recording

What you will need

A Seeed XIAO ESP32-S3 Sense - the "Sense" expansion carries the OV2640 camera and the PDM microphone, and the module has the 8 MB PSRAM the camera frame buffers need.
The arduino-esp32 3.x core (it ships the esp_camera and ESP_I2S components this example uses). On any other board / core the sketch still compiles, but the media routes are replaced by a short "needs a XIAO ESP32-S3 Sense" message so it builds everywhere.

The camera and microphone are wired on the board; the pins are already set in the sketch:

Signal	GPIO
Camera XCLK	10
Camera SDA/SCL	40 / 39
Camera VSYNC/HREF/PCLK	38 / 47 / 13
Camera data Y2-Y9	15, 17, 18, 16, 14, 12, 11, 48
Mic PDM CLK/DATA	42 / 41

Configure and flash

Open MediaStreaming.ino and set the two CHANGE ME lines to your Wi-Fi SSID / PASSWORD.
In the Arduino IDE select XIAO_ESP32S3 and set PSRAM: "OPI PSRAM" (the camera fails to start without it). Flash, then open Serial Monitor @ 115200; you will see:

camera OK, mic OK

Open http://192.168.1.232

Browse to that address - the page shows live video with an audio player, or open /video and /audio.wav directly.

With arduino-cli the same build is:

arduino-cli compile --upload -p <PORT> \
  --fqbn esp32:esp32:XIAO_ESP32S3:PSRAM=opi \
  --library . examples/L7-Application/MediaStreaming

Verify without a browser

curl -s http://<BOARD_IP>/photo.jpg  -o photo.jpg          # a JPEG (starts FF D8, ends FF D9)
curl -s http://<BOARD_IP>/clip.wav   -o clip.wav           # a 2 s WAV (RIFF/WAVE, 16 kHz PCM)
curl -s -m 5 http://<BOARD_IP>/video -o video.mjpeg        # several JPEG frames + --detframe boundaries
curl -s -m 5 http://<BOARD_IP>/audio.wav -o audio.wav      # ~160 KB in 5 s = the 32 KB/s real-time rate

On real hardware this streams ~21 fps VGA video and real-time audio; the camera's JPEG rate and the Wi-Fi link set the pace, not the send path. See FEATURE_PERFORMANCE.md for the measured numbers.

Build note (PlatformIO)

The example uses only the base server (no PC_ENABLE_* flag), so a plain compile check needs no build flags:

pio ci examples/L7-Application/MediaStreaming --board esp32-s3-devkitc-1 --lib "."

On a non-S3 board this compiles the web-only stub; build it for the XIAO ESP32-S3 Sense (PSRAM on) to get the camera and microphone.