Multi-camera video synchronization using inaudible acoustic markers — no timecode hardware, no clapper board, no HDMI sync cable. You press a button; every camera in the room hears a short ultrasonic tone burst; post-processing aligns the recordings by matching the tone.

cheepsync-mark emits a ~200ms FSK-encoded burst at 18 kHz (adjustable) when you press a button. Each press broadcasts a unique marker ID (0–255 per session) and logs the exact wall-clock time. Think of it as a digital slate clap.

cheepsync-sync finds those markers in your recorded video files and either:

• --find — prints a table of which markers appear in which files and when
• default — prepends blank video to earlier-starting cameras and low-pass filters the sync tone out of all tracks, producing aligned files you can drop into any editor

• Python 3.9+
• ffmpeg and ffprobe on PATH (for cheepsync-sync)
• A speaker within ~3 m of all cameras (Pimoroni Pirate Audio 3W hat works well on the Pi)

pip install cheepsync           # mac/linux
pip install cheepsync[rpi]      # Raspberry Pi (adds gpiozero)

1.  Start recording on all cameras.
2.  Press the cheepsync-mark button (or spacebar) — one press per sync point.
3.  Stop cameras.
4.  Run cheepsync-sync to align the files.

Pressing the button multiple times creates redundant sync points — useful if one recording was too noisy or the signal was blocked. You only need one common marker between cameras.

Plays a marker burst on each button press and writes a session log.

cheepsync-mark [OPTIONS]

Trigger:

• Raspberry Pi — physical button on BCM 5 (Pimoroni Pirate Audio A button, detected automatically via gpiozero)
• Mac / Linux — press SPACEBAR; Q or Ctrl+C to end session

Session log — written to the current directory as cheepsync_YYYYMMDD_HHMMSS.json:

{
  "session_id": "20260406_142315",
  "started": "2026-04-06T14:23:15.000Z",
  "base_freq": 18000.0,
  "fsk_shift": 500.0,
  "markers": [
    {"id": 0, "time": "2026-04-06T14:23:45.123Z"},
    {"id": 1, "time": "2026-04-06T14:24:12.456Z"}
  ]
}

Detects markers in video files and produces aligned output.

cheepsync-sync [OPTIONS] VIDEO [VIDEO ...]

cheepsync-sync cam_a.MOV cam_b.MOV --find

Prints a table of every detected marker, which files contain it, and the time it appears in each:

  Marker  cam_a.MOV     cam_b.MOV
  ------  ----------    ----------
       0 ✓  10.000s      6.300s
       1 ✓  25.000s     21.300s
       3    89.001s         —

  2 of 3 marker(s) present in all files.

Pass --session-log cheepsync_*.json to annotate the table with wall-clock times.

cheepsync-sync cam_a.MOV cam_b.MOV --output-dir ./synced

1. Detects markers in each file
2. Picks the common markers and computes per-camera time offsets
3. Generates a _START offset clip for each non-reference camera
4. Place each _START clip before its video on the NLE timeline to align

Optionally pass --lowpass-hz 17500 to also generate _filtered copies with the sync tones removed from audio.

Each marker burst is 200 ms total:

[30ms sync pulse @ base_freq] [10ms silence] [8 FSK bits × 20ms each]

The 30ms sync pulse followed by 10ms of silence acts as a structural header: the FSK decoder skips past it to find the 8 data bits that follow.

Detection pipeline:

1. Goertzel filter at base_freq in 5ms windows — computes magnitude at the exact sync tone frequency (single DFT bin, no broadband pre-filter needed)
2. Threshold onset detection — Goertzel magnitude must exceed noise_floor × threshold_factor, with a 250ms refractory period between detections
3. FSK decode — at each onset, skip the sync pulse + gap, then compare Goertzel magnitudes at the two FSK frequencies across 8 bit windows

18 kHz is the default — above the hearing range of most adults, below the recording cutoff of common consumer cameras. Some cameras roll off above 15–16 kHz.

If markers aren't detected: open a recording in Audacity, switch to Spectrogram view (range 14–22 kHz), and look for a short bright stripe at 18 kHz at the moment of the button press. If no stripe is visible, lower --base-freq in 500 Hz steps until it appears.

If you hear the tone: lower --base-freq toward 18000–19000 Hz to move it higher. Young ears can sometimes hear above 17 kHz; test with your youngest participants.

If false positives persist: try a different --base-freq that doesn't coincide with camera electronics noise. Common interference bands: 15.6 kHz (CRT horizontal scan, rare now), 17–19 kHz (some LED drivers), 20 kHz (some camera DSPs).

pip install cheepsync[rpi]
cheepsync-mark --device "sndrpihifiberry"   # find exact name with --list-devices

The MAX98357A I²S DAC on the Pirate Audio hat appears as an ALSA device. Use --list-devices to find its name on your system. Output is always stereo (L+R identical) as required by the I²S DAC.

The A button (BCM 5) is used automatically when gpiozero is installed. No configuration needed.