okayvid/mirror-mirror

An open source voice assistant with an in-depth visual dashboard

Mirror Mirror

A program to run a voice assistant with a graphical status display

NOTE: currently undergoing major rework, will be rebranded after that

I created this to be run on a raspberry pi magic mirror and be like mirror mirror on the wall from shrek/snow white, hence the name.

Details

The program uses pocketsphinx to listen for voice commands, while in the background running a graphical status display (using threading) with the time/weather. The voice assistant also uses the graphical display to show relevant information for the command the user used.

Commands

• time
• date
• weather
• joke
• general greeting
• news
• definitions

The assistant listens for keywords, not phrases, for example:

• "assistant tell me a joke" - this will work
• "assistant tell a joke" - this will also work

The default wake word is "assistant"; this can be changed on lines 59 and 60 of main.py.

More coming soon!

Dependencies

• vosk
• pysimplegui (Tkinter port)
• gTTS
• requests
• playsound
• sounddevice
• websockets

Required API key files

You will need to create a couple of files in a directory called Keys/. These are:

• openweathermap_key.json:

  {"key": "INSERT OPENWEATHERMAP API KEY", "lat": "INSERT YOUR LOCATION'S LATITUDE", "lon": "INSERT YOUR LOCATION'S LONGITUDE"}

• newsapi_key.txt:

  <insert api key here>

Important: Starting VOSK server

This project uses the VOSK speech recognition API, and requires a VOSK server to be running on port 2700.

• For x86/amd64: You can run the VOSK websocket server using docker: docker run -d -p 2700:2700 alphacep/kaldi-en:latest

• For ARM (also works on x86/amd64):

  1. Clone the vosk server repository https://github.com/alphacep/vosk-server/
  2. Download a language model https://alphacephei.com/vosk/models - I'm running this project on a raspberry pi 4, for which I found the en-us small model worked best - I didn't have enough RAM to use the normal sized model. Otherwise however the normal model is useful as it is slightly more accurate.
  3. Extract the model into a directory called "model" which is itself in the vosk-server/websocket/ directory. (alternatively put it in the vosk-server/websocket-cpp/ folder if you are going to run the CPP version of the server)
  4. Run the server - either asr_server.py if you're using the python version in vosk-server/websocket/, or asr_server.cpp if you're using the CPP version in vosk-server/websocket-cpp/ directory.