fwcd / okpi

Virtual assistant with offline voice recognition for Raspberry Pi

OkPi

Virtual assistant with offline voice recognition. Despite being primarily designed for Raspberry Pi, it should compile on any computer running Linux or macOS.

Installation

Pocket Sphinx

• The PocketSphinx installation currently requires Node.js 6
  • This can be achieved by using a version manager such as nvm:
    • nvm install 6
• Install the PocketSphinx dependencies:
  • apt-get install cmake pkg-config libpcre3-dev bison
  • swig from GitHub
  • sphinxbase from GitHub
  • pocketsphinx from GitHub
  • Note that certain tweaks might be required prior to install:
    • export PKG_CONFIG_PATH=/usr/local/lib/pkgconfig
    • ln -s /usr/bin/swig3.0 /usr/bin/swig (Create a symlink for swig3.0)
  • npm install cmake-js -g

Say.js

• Install Festival with a default voice:
  • apt-get install festival festvox-rablpc16k

NPM Dependencies

• npm install

Launching

• Make sure that this repository (OkPi) is in the same folder as the cloned pocketsphinx repository
• npm run start
  • Or alternatively using watch mode:
  • npm run watch
  • npm run launch (in a separate terminal)

Architecture

• The application roughly implements the following structure:

• Speech input

  • -> AudioInput
  • -> SpeechRecognitionEngine
  • -> UtteranceProcessor
  • -> Skill
  • -> OutputFacade

okpi's Issues

Command Line Arguments

Implement configuration options (such as the language model or the keyphrase) as CLI arguments.

Plugin API

Create a plugin/extension system that allows the user to add custom "skills".