Cubie-3 Code

The full version controlled code for this project is available at Cubie-3 GitHub Respository.

Note

This code requires the Sunfounder Fusion HAT+ Python library. You can find installation instructions in the Sunfounder Fusion HAT+ - What Can’t This Thing Do? video.

Example code to drive the motors:

from cubie_3 import Cubie3
import asyncio

async def main():
    robot = Cubie3()

    # Move forward at 30% speed
    await robot.forward(0.3)
    await asyncio.sleep(1)

    # Strafe right
    await robot.strafe_right(0.5)
    await asyncio.sleep(0.5)

    # Stop all motors
    await robot.stop()

asyncio.run(main())

Voice Recognition with Vosk

Cubie-3 uses Vosk for offline speech-to-text recognition. This means your robot can respond to voice commands without needing an internet connection - all processing happens locally on the Raspberry Pi.

What is Vosk?

Vosk is a lightweight, offline-capable speech recognition engine that:

• Works without an internet connection
• Supports 20+ languages
• Runs on resource-constrained devices like Raspberry Pi
• Has small language models (as small as 50MB)

How It Works in Cubie-3

The voice recognition system uses the Sunfounder Fusion HAT+ library which wraps Vosk functionality. When you call the listen() method:

1. Audio capture - The USB microphone captures audio continuously
2. Real-time processing - Vosk processes audio in a stream, providing partial results as you speak
3. Command matching - When a final result is recognized, Cubie-3 checks if it matches a known command
4. Action execution - Matching commands trigger the corresponding robot movement

Setting Up Voice Recognition

1. Hardware Requirements

• USB microphone (essential for clear audio input)
• Sunfounder Fusion HAT+ installed on your Raspberry Pi

2. Configure Your Microphone

First, identify your USB microphone:

arecord -l

You’ll see output like:

card 1: Device [USB Audio Device], device 0: USB Audio [USB Audio]

Test recording (replace 1,0 with your card and device numbers):

arecord -D plughw:1,0 -f S16_LE -r 16000 -d 3 test.wav
aplay test.wav

If the recording is too quiet or muted, adjust levels:

alsamixer

• Press F6 to select your USB microphone
• Find the Mic or Capture channel
• Ensure it shows [OO] (unmuted) - press M to toggle
• Use arrow keys to adjust volume

3. Install Dependencies

The Fusion HAT+ library handles Vosk installation, but if you need to install manually:

sudo apt install portaudio19-dev python3-pyaudio
pip3 install vosk

The language model downloads automatically on first use.

Voice-Controlled Demo

Here’s the complete voice control example from the Cubie-3 repository:

from cubie_3 import Cubie3
import asyncio

async def main():
    cubie3 = Cubie3()

    # Announce startup
    cubie3.tts.say("Starting movement demo.")

    # Start listening for voice commands
    await cubie3.listen()

if __name__ == "__main__":
    asyncio.run(main())

Recognized Voice Commands

Cubie-3 responds to these voice commands:

How the Listen Method Works

The listen() method in the Cubie3 class implements continuous speech recognition:

# Simplified example of how listen() works internally
from fusion_hat.stt import Vosk as STT

stt = STT(language="en-us")

while True:
    print("Listening...")
    for result in stt.listen(stream=True):
        if result["done"]:
            # Final recognized text
            command = result['final'].lower()
            if "forward" in command:
                await robot.forward(0.5)
            elif "stop" in command:
                await robot.stop()
            # ... more commands
        else:
            # Partial result (updates as you speak)
            print(f"Heard: {result['partial']}", end="\r")

Text-to-Speech (TTS)

Cubie-3 can also speak back to you using the say() method:

from cubie_3 import Cubie3

robot = Cubie3()

# Robot speaks using Espeak TTS
robot.tts.say("Hello, I am Cubie-3!")
robot.tts.say("Ready for your commands.")

The TTS system uses Espeak, which is lightweight and built into Raspberry Pi OS. While the voice sounds robotic, it’s highly configurable for volume, pitch, and speed.

Tips for Better Voice Recognition

1. Microphone position - Keep the microphone 15-30 cm from your mouth
2. Quiet environment - Background noise reduces accuracy
3. Clear speech - Speak clearly and at a normal pace
4. Pause between commands - Give the system time to process

Troubleshooting

“No audio detected”

• Check microphone connection with arecord -l
• Verify volume levels in alsamixer
• Test recording with arecord command

“Commands not recognized”

• Speak the exact command phrases
• Check you’re using the correct language model
• Try speaking more slowly and clearly

“Model download fails”

• Check internet connection (only needed for initial download)
• Manually download from Vosk Models
• Use the small English model: vosk-model-small-en-us-0.15

Supported Languages

Vosk supports many languages. Change the language when initializing:

# For British English
stt = STT(language="en-gb")

# For German
stt = STT(language="de")

# For French
stt = STT(language="fr")

Available language codes include: en-us, en-gb, de, fr, es, it, pt, ru, cn, ja, ko, and many more.

Further Resources

• Vosk Official Documentation
• Vosk GitHub Repository
• Sunfounder Fusion HAT+ STT Documentation
• Vosk Language Models