Voix

This package uses the Google Text-To-Speech API to allow for seamless human robot interaction by sending navigation commands issued by the user to robot. Main operating system used was ROS while Turtlebot and Gazebo were used for simulation. Actual experiments were conducted using RosAria and the Pioneer3AT robot.

Academic Citations

Please cite this IEEE publication when using this package for any academic projects: M. Elmzaghi, M. Fahad and Y. Guo, "Implementing Robust Voice-Control for Human Robot Interaction for Autonomous Robot Guides," 2019 IEEE MIT Undergraduate Research Technology Conference (URTC), Cambridge, MA, USA, 2019, pp. 1-6, doi: 10.1109/URTC49097.2019.9660499.

Why use Google API?

The goal is to create natural human robot interaction. Current voice processing software requires users to design speech recognition algorithms, collect datasets, as well as understand Natural Language Processing methods. The accuracy of these methods still does not exceed that of Google Cloud API given its massive datasets from the various voice services as well as the use of deep learning neural networks. By incorporating an accurate and robust voice processing software easily integrated with popular robotics software, users can include robust voice control into their robotics projects.

Installation

To use this software package, ensure you are running ROS for your respective system as well as have installed Google Cloud SDK (which is detailed below.) However, we need to install PortAudio so we can use PyAudio to get mic data.

sudo apt-get install portaudio19-dev

Install all the requirements using pip by cloning the Github repo and installing all the packages in requirements.txt.

cd ~/catkin_ws/src
git clone https://github.com/moeelm/voice_ros.git
cd voice_ros
pip install -r requirements.txt

Google Cloud Setup

Follow the instructions here for configuring your Google Cloud project and installing the SDK for authentication. You will need a google/gmail account.

Usage of the Google Cloud SDK requires authentication. This means you require an API key and an activated service account to utilize the APIs.

1. Setup a service account
2. Download the service account key as a JSON.
3. Check you have GOOGLE_APPLICATION_CREDENTIALS in your environment. This should be the path to the keys.

   export GOOGLE_APPLICATION_CREDENTIALS='/path/to/key'

4. Run the authentication command:

   gcloud auth activate-service-account --key-file GOOGLE_APPLICATION_CREDENTIALS

Usage

Follow the steps below to setup the package properly.

Configure topics

Go into the config directory and change the following parameters in the params.yaml file:

• results_topic: (Optional) The topic where your results will be published.
• project_id: The name of your project for the Google Speech node. This is the name of your Google Cloud project when going through the Google Cloud setup.

Launching nodes

To start the voice processing node, run the following command:

roslaunch voice_ros voice.launch

Autonomous Navigation Using Turtlebot

To run autonomous navigation using the Turtlebot robot as well as RVIZ simulator,

1. Launch Turtlebot as well as Gazebo environment

   roslaunch turtlebot_gazebo turtlebot_world.launch

2. Run the map

   rosrun map_server map_server /PATH/TO/myMap.yaml

3. Run the navigation demo. AMCL (Adaptive Monte Carlo Localization) is used for localizing the robot

   roslaunch turtlebot_gazebo amcl_demo.launch

4. Launch ROS Vizualization (RVIZ)

   roslaunch turtlebot_rviz_launchers view_navigation.launch

   The 2D Nav Goal sends a navigation goal, 2D Pose Estimate sets the pose of the robot.

ROS Nodes

mic_client

ROS node receives text from the Google Cloud Speech API and publishes it onto text_topic (see config/params.yaml). This is parsed and sent to the navigation node, nav_v1.py. single_utterance is set to true to allow for single user input. This prevents the voice processing from being continous, allowing the intent to be parsed and detected.

Published Topics

text_topic (std_msgs/String) Acquired text from the Google Cloud Speech API. text2_topic(std_msgs/String) Intent detected in voice command.

Navigation

ROS node that takes text from the mic_client node, publishes the parsed intent and sends the spatial location associated with the intent to the \move_base_simple/goal topic. For example, 'microwave'has the position coordinates (2.00, -5.00, 0.000) mapped to it.