Installation

1. Hardware dependencies

   1. Microphone: You must have a good microphone. Typical computer/notebook microphones are not sufficient.

      • For example, the Sony PlayStation 3 Eye camera.

   2. Loud speaker, obviously.

   3. Normal desktop CPU.

      • We'll be working on Raspberry Pi 4 support soon. Raspberry Pi 3 will not be supported.

2. Install OS dependencies

   • Ubuntu, Debian etc.

     • # apt install yarnpkg mpg123 mpd sox libasound2-dev default-jre-headless default-jdk git

   • Fedora, RedHat etc.

     1. Install RPM Fusion (for mpd)

        • # dnf install https://download1.rpmfusion.org/free/fedora/rpmfusion-free-release-$(rpm -E %fedora).noarch.rpm

     2. # dnf install mpg123 mpd sox alsa-lib-devel java-1.8.0-openjdk-headless java-1.8.0-openjdk-devel git

     3. Install Yarn

        1. Complete step 3 below.

        2. $ npm install -g yarn

   • Mac

     1. # brew node yarn sox mpd mpg123 openjdk git

     2. Skip step 3.

3. Install node.js

   1. Go to the node.js website and download node version 14 or later.

      • node version 12 may work.

      • node version 10 in Ubuntu 20.04 is too old.

   2. Extract the archive somewhere.

   3. ln -s/your/path/to/node//usr/local/bin/node

4. Install MaryTTS

   1. git clone https://github.com/marytts/marytts/

   2. Build it with ./gradlew

   3. Install voices by runningwget `grep location.*cmu-slt download/marytts-components.xml|sed -e "s|.*href=\"||" -e "s|\".*||"` unzip voice-cmu-slt-5.2.zip or alternatively starting the UI app./gradlew runInstallerGui

      • en-US cmu-slt

      • en-US cmu-rms

      • en-GB dfki-spike

      • en-GB dfki-prudence

   4. Restart the server: ./gradlew run

   5. Open http://localhost:59125 in a browser, to check whether the server is running and the voices are listed in the dropdown.

5. Download data files

   1. Download the Pia app data files

   2. cd to the Pia source directory

   3. mkdir data

   4. cd data/

   5. tar xjf /path/to/pia-data.tar.bz2

   6. mkdir deepspeech-0.7.3-models

      • The DeepSpeech model version needs to match the version of DeepSpeech npm module perfectly. Double-check the version that you have with grep version node_modules/deepspeech/package.json.

   7. cd deepspeech-0.7.3-models/

   8. Download the DeepSpeech 0.7.3 model file and DeepSpeech 0.7.3 scorer file into this directory.

   9. They are 1.2 GB, so while they download, you can already adjust config.json, see step 6.3. below.

6. Install node packages

   1. cd to the Pia source directory

   2. yarn install

   3. cp config-min.json config.json

   4. Adapt config.json

7. Run Pia

   1. cd to the Pia source directory

   2. yarn start

   3. You should see

      (...)
      Applications loaded:
        clock: success
      (...)
      Listening to your command...

   4. Say "Hey Edison, what time is it?"

   5. Find more commands to say by looking at app/*/intents.en.json

8. Configure the applications

9. Create your own voice apps :-)

   1. Implement your voice commands in JavaScript, with the help of npm modules.