Hotword

Hotword detection for the PinguimBots robot Theta. Hotword can also be called wakeword.

Why Porcupine

Porcupine stands out among the tested libraries below. It is fast in detecting the spoken wakeword.

Tested python libraries

• SnowBoy (da KITT.AI)
  • does not exist anymore
• PocketSphinx da CMUSphinx
  • not very accurate
  • offline
  • open source
  • Pypi: https://pypi.org/project/pocketsphinx/
• Pvporcupine
  • needs AcessKey
  • only three models per month
  • only 3 computers to use a model per month
    • needs to train another ppn every month
    • Two types:
    • Porcupine wake word
      • fast detection
      • efficient
      • Source: https://picovoice.ai/docs/quick-start/porcupine-go/
    • Rhino Speech-to-Intent
      • can only make 10 commands in free version
      • Source: https://picovoice.ai/platform/rhino/
• EfficientWord-Net
  • needs to train a model for specific words
    • models for words like Alexa, siri, Mycrosoft
    • Source code: https://github.com/Ant-Brain/EfficientWord-Net
• Mycroft Precise
  • requires python 3.6 until 3.9
  • not precise
    • Research paper: https://worldscientific.com/doi/10.1142/S0219649222500599
    • https://ant-brain.github.io/EfficientWord-Net/
    • Source code: https://github.com/MycroftAI/mycroft-precise

Install requirements

$ pip install -r requirements.txt

It also requires the archive din-ding.mp3 and the ppn file from the picovoice console.

How to run

Please read Notes below before you run this code.

$ git clone https://github.com/pinguimbotsathome/Hotword
$ cd Hotword
$ python porcupine.py

How the code works

It uses pvrecorder to open the mic and start listening to the audio stream. When it hears the wakeword "Hey Theta", it prints Detected on the screen and plays the audio file din-ding.mp3.

Notes

This instructions are crucial to run the script:

1. Access key => in code: key => key from picovoice account. Create an account in picovoice and copy the Access key.
2. keyword_p => path to ppn file. Go to picovoice console and train the wakeword "Hey Theta". Choose Linux as device and download the file. Put the file inside the Hotword directory.
3. sensitivities => It is a number between 0 and 1. Deafult is 0.55.A higher sensitivity results in fewer misses at the cost of increasing the false alarm rate.
4. device_index => the default number is -1. Change it to match your device input. Usually is 1.