Build a Talking Skull with a Servo Jaw and Glowing Eyes

A skull on the porch that opens its mouth and says "Welcome to my haunted house" is the kind of prop people remember. It is also a tidy lesson in timing. In this post you will build a talking skull on an Arduino Uno: a motion sensor wakes it, green LED eyes glow, a servo moves the jaw, and a DFPlayer Mini plays the voice. The part worth learning is how to keep the jaw in step with the sound.

How the pieces fit

• A PIR sensor watches for movement. Its output goes high when someone walks by.
• A DFPlayer Mini is a small MP3 module. You tell it over a serial link to play a numbered file from its microSD card, and its BUSY pin stays low while a track plays.
• A speaker on the DFPlayer's SPK_1 and SPK_2 pins makes the sound.
• A servo moves the jaw. Two green LEDs are the eyes.

One honest note up front: the simulated DFPlayer makes no sound. It shows which file it is playing and how far into it, so you can see the track start and end on time. The audio is for the real speaker.

What you need

• An Arduino Uno
• A DFPlayer Mini and a small speaker
• A servo
• A PIR motion sensor
• Two green LEDs, two 220 ohm resistors and one 1 k resistor
• Jumper wires

Wire it up

Part and pin	Uno pin	Notes
PIR VCC, GND	5V, GND	Power.
PIR OUT	2	High while it sees movement.
DFPlayer VCC, GND	5V, GND	Power.
DFPlayer TX	10	The Uno listens here.
DFPlayer RX	11, through the 1 k resistor	The Uno talks here.
DFPlayer BUSY	3	Low while a track plays.
DFPlayer SPK_1, SPK_2	Speaker pins 1 and 2	The sound.
Servo PWM	9	The jaw.
Servo VCC, GND	5V, GND	Power.
Left eye	5, through 220 ohms, to GND	Anode toward the resistor.
Right eye	6, through 220 ohms, to GND	Same.

If you would rather skip the wiring, the built-in Talking skull project has all of this laid out on a breadboard and opens ready to run. Pins 5 and 6 both have PWM, so the eyes can glow dimly as well as brightly.

The complete sketch

This is a trimmed version of the built-in project, with the same pins.

#include <SoftwareSerial.h>
#include <DFRobotDFPlayerMini.h>
#include <Servo.h>

const int PIR_PIN = 2;
const int BUSY_PIN = 3;
const int EYE_LEFT = 5;
const int EYE_RIGHT = 6;
const int JAW_PIN = 9;
const int JAW_SHUT = 5;                 // degrees
const unsigned long REST_MS = 10000UL;  // quiet time between visitors

struct Cue {
  unsigned int at;     // milliseconds from the start of the track
  unsigned char angle; // jaw angle in degrees
};

// "Wel-come... to my HAUN-TED house!"
const Cue SCRIPT[] = {
  {0, 30},    {200, 5},   {320, 35},  {560, 5},
  {900, 20},  {1040, 5},  {1120, 25}, {1260, 5},
  {1450, 45}, {1700, 10}, {1780, 40}, {1980, 5},
  {2100, 35}, {2450, 5},
};
const int CUES = sizeof(SCRIPT) / sizeof(SCRIPT[0]);

SoftwareSerial link(10, 11);  // RX on 10, TX on 11
DFRobotDFPlayerMini player;
Servo jaw;

bool haveSound = false;
bool talking = false;
unsigned long startedAt = 0;
unsigned long quietSince = 0;
int nextCue = 0;

void eyes(int level) {
  analogWrite(EYE_LEFT, level);
  analogWrite(EYE_RIGHT, level);
}

void startTalking() {
  talking = true;
  startedAt = millis();
  nextCue = 0;
  eyes(255);
  if (haveSound) {
    player.play(1);
  }
  Serial.println("Visitor! Talking.");
}

void setup() {
  Serial.begin(9600);
  link.begin(9600);
  pinMode(PIR_PIN, INPUT);
  pinMode(BUSY_PIN, INPUT);
  jaw.attach(JAW_PIN);
  jaw.write(JAW_SHUT);
  eyes(20);

  haveSound = player.begin(link);
  if (haveSound) {
    player.volume(24);
  } else {
    Serial.println("No DFPlayer answered. The skull will mime.");
  }
  Serial.println("Skull ready. Saying hello once.");
  startTalking();
}

void loop() {
  unsigned long now = millis();

  if (talking) {
    unsigned long t = now - startedAt;

    while (nextCue < CUES && t >= SCRIPT[nextCue].at) {
      jaw.write(SCRIPT[nextCue].angle);
      nextCue++;
    }

    // BUSY is LOW while a track plays. Done when the script is spent
    // and the player says the track is over.
    bool playing = haveSound && digitalRead(BUSY_PIN) == LOW;
    if (nextCue >= CUES && !playing && t > 500) {
      talking = false;
      quietSince = now;
      jaw.write(JAW_SHUT);
      eyes(20);
      Serial.println("Done. Waiting for the next visitor.");
    }
    return;
  }

  if (now - quietSince >= REST_MS && digitalRead(PIR_PIN) == HIGH) {
    startTalking();
  }
}

I compiled this sketch for the Uno with Mokxi's own build path and ran it in the simulator on this wiring.

How the jaw keeps time

There are two honest ways to sync a jaw to a voice. One is to follow the audio: rectify and smooth the speaker signal, read it on an analog pin, and open the jaw as far as the voice is loud. That works with any recording, but the jaw chatters on every breath.

The other is to script it. You listen to the line once, note when each syllable starts, and write those moments down as cues. It takes a few minutes per line and looks crisp, because you decide exactly when the mouth shuts. This sketch is scripted.

Each Cue is a time in milliseconds from the start of the track and a jaw angle. {0, 30}, {200, 5} means "at the very start open to 30 degrees, then at 200 milliseconds close to 5". Read down the list and you can see the syllables: open, close, open, close.

Line by line

• The three #include lines bring in the serial link, the DFPlayer library and the servo library, all built in under their usual names.
• JAW_SHUT = 5 is the closed angle in degrees. REST_MS is the quiet time between visitors.
• SCRIPT[] is the cue list. CUES counts the entries.
• SoftwareSerial link(10, 11) makes a second serial port on pins 10 and 11, because the Uno's main serial is busy with the monitor.
• eyes(level) sets both eyes with analogWrite.
• startTalking() records the start time, resets the cue counter, turns the eyes up to 255 and tells the player to play track 1.
• In setup(), player.begin(link) returns false if no DFPlayer answers. We store the result in haveSound, so the skull can still mime the jaw without sound.
• In loop(), t = now - startedAt is how far into the track we are.
• The while loop is the heart of it. It fires every cue whose moment has come, moving the jaw to that cue's angle, then steps to the next one. If the sketch is a little late, it catches up instead of skipping.
• digitalRead(BUSY_PIN) == LOW asks the player whether the track is still playing.
• The skull finishes only when the cues are spent and BUSY says the track is over. That way a longer recording never leaves the jaw hanging open.
• The last if waits out the rest period, then talks again when the PIR reads high.

What you should see

Press Run. The skull says hello once at power-up, so you know the speaker, the card and the servo all work. In my run the serial monitor printed Skull ready. Saying hello once., then Visitor! Talking., then Done. Waiting for the next visitor.

While it talks, the servo steps through the cue list, the eyes go bright, and the DFPlayer part shows a status like playing 0001.mp3 with a clock. When the track ends the eyes dim and the jaw closes. Click the PIR to wake it again after the ten second rest.

On a real porch

Give the servo its own 5 volt supply with its ground joined to the Uno's, since a servo can pull enough current to reset the board. Most PIR modules need about a minute after power-up to settle and fire at random until they do.

Try next

1. Record your own line, time each syllable and replace SCRIPT.
2. Make the eyes flare on wide-open syllables, using the cue angle as a stand-in for loudness.
3. Add a second track and pick one at random each visit.
4. Read the Halloween props guide for four more builds, and the servo part page for its pins.

Try it on Mokxi

Open the Talking skull project to run the full build, or start from the Arduino simulator and place the parts yourself. The Arduino Uno board page lists the pins used here.

Ready to build it yourself? Create your free Mokxi account and open the editor in your browser. No installs, no parts to buy.