SSH into your Pi 5 or work directly on it

Assuming fresh Debian 12 install

============================================

STEP 1: System Update & Essential Tools

============================================

sudo apt update
sudo apt upgrade -y

Install build essentials and development tools

sudo apt install -y \
build-essential \
git \
wget \
curl \
vim \
python3 \
python3-pip \
python3-venv \
python3-dev

Verify Python version (should be 3.11.x)

python3 –version

============================================

STEP 2: Audio System Setup

============================================

Install ALSA and PortAudio (required for pyaudio)

sudo apt install -y \
alsa-base \
alsa-utils \
portaudio19-dev \
libasound2-dev \
libportaudio2 \
libportaudiocpp0

Test audio devices are detected

arecord -l # List recording devices
aplay -l # List playback devices

You should see your USB audio interface listed

If you don’t see your USB device, plug it in and run again

Test microphone recording (Ctrl+C to stop, then play it back)

arecord -D hw:0,0 -f cd test.wav

Press Ctrl+C after 3 seconds

aplay test.wav

If you hear your voice, mic works. If not, try different hw:X,Y values

============================================

STEP 3: Create Project Directory & Virtual Environment

============================================

Create project directory

mkdir -p ~/medic-sam
cd ~/medic-sam

Create Python virtual environment

python3 -m venv venv

Activate virtual environment

source venv/bin/activate

Your prompt should now show (venv) prefix

Always activate this before working: source ~/medic-sam/venv/bin/activate

Upgrade pip in the virtual environment

pip install –upgrade pip

============================================

STEP 4: Install Python Dependencies

============================================

Install audio processing libraries

pip install pyaudio

If pyaudio fails, you might need:

sudo apt install python3-pyaudio

Then: pip install pyaudio

Install OpenWakeWord and its dependencies

pip install openwakeword

This will install:

- onnxruntime (inference engine)

- numpy (numerical operations)

- scipy (audio processing)

- resampy (audio resampling)

Install additional dependencies we’ll need later

pip install \
numpy \
scipy

Verify installations

python3 -c “import pyaudio; print(‘PyAudio:’, pyaudio.version)”
python3 -c “import openwakeword; print(‘OpenWakeWord installed’)”
python3 -c “import numpy; print(‘NumPy:’, numpy.version)”

============================================

STEP 5: Download OpenWakeWord Models

============================================

OpenWakeWord models are downloaded automatically on first run,

but we can pre-download them to verify everything works

python3 << EOF
from openwakeword.model import Model
print(“Downloading default wake word models…”)
oww = Model()
print(“Available models:”, list(oww.models.keys()))
print(“Models downloaded successfully!”)
EOF

You should see models like: hey_mycroft, alexa, hey_jarvis, timer

============================================

STEP 6: Audio Permissions & Configuration

============================================

Add your user to the audio group (if not already)

sudo usermod -a -G audio $USER

You’ll need to log out and back in for this to take effect

Or run: newgrp audio

Verify audio group membership

groups

Should show ‘audio’ in the list

============================================

STEP 7: Create Test Script

============================================

Create the wake word test script

cat > wake_word_test.py << ‘SCRIPT_EOF’

!/usr/bin/env python3

”“”
Medic Sam - Wake Word Detection Test
Minimal script to verify OpenWakeWord is working
“”“

import pyaudio
import numpy as np
from openwakeword.model import Model

Audio configuration

CHUNK_SIZE = 1280 # OpenWakeWord expects 16kHz audio in chunks of 1280 samples (80ms)
SAMPLE_RATE = 16000
CHANNELS = 1
FORMAT = pyaudio.paInt16

def main():
print(“Initializing Medic Sam Wake Word Detection…”)
print(“=” * 50)

# Initialize OpenWakeWord model with default wake words
oww_model = Model(
    wakeword_models=["hey_mycroft"],
    inference_framework='onnx'
)

print(f"Loaded models: {list(oww_model.models.keys())}")
print("Say 'Hey Mycroft' to trigger detection")
print("(We'll replace this with custom 'Medic Sam' model later)")
print("Press Ctrl+C to exit")
print("=" * 50)

# Initialize PyAudio
audio = pyaudio.PyAudio()

# Find and display available input devices
print("\nAvailable audio input devices:")
for i in range(audio.get_device_count()):
    info = audio.get_device_info_by_index(i)
    if info['maxInputChannels'] > 0:
        print(f"  [{i}] {info['name']}")

# Open audio stream
stream = audio.open(
    format=FORMAT,
    channels=CHANNELS,
    rate=SAMPLE_RATE,
    input=True,
    frames_per_buffer=CHUNK_SIZE
)

print("\nListening for wake word...\n")

try:
    while True:
        # Read audio chunk
        audio_data = stream.read(CHUNK_SIZE, exception_on_overflow=False)

        # Convert to numpy array (int16)
        audio_array = np.frombuffer(audio_data, dtype=np.int16)

        # Get prediction from OpenWakeWord
        prediction = oww_model.predict(audio_array)

        # Check if wake word detected (threshold: 0.5)
        for model_name, score in prediction.items():
            if score > 0.5:
                print(f">>> WAKE WORD DETECTED! (confidence: {score:.2f})")
                print(f"    Model: {model_name}")
                print()

except KeyboardInterrupt:
    print("\n\nShutting down...")

finally:
    # Clean up
    stream.stop_stream()
    stream.close()
    audio.terminate()
    print("Audio stream closed. Goodbye!")

if name == “main”:
main()
SCRIPT_EOF

Make it executable

chmod +x wake_word_test.py

============================================

STEP 8: Test the Setup

============================================

echo “”
echo “============================================”
echo “Setup complete! Now testing…”
echo “============================================”
echo “”

Run the test script

python3 wake_word_test.py

You should see:

1. List of available audio devices

2. “Listening for wake word…” message

3. Say “Hey Mycroft” - should print “WAKE WORD DETECTED!”