voice_recognition

Node to perform keyword detection on an audio stream, and publish recognised words on a ROS topic.

Installation

1. Clone this repo to your catkin workspace src folder, and change to the voice_recognition directory.

2. Install dependencies. Run bin/check_deps.sh to see which packages are required.

3. Sox can be installed from apt: sudo apt install sox

4. The python packages can be installed with pip (which can be installed with sudo apt install python3-pip). In the steps below, choose either tensorflow or tensorflow-gpu depending on your device. CUDA and CUDNN must be properly setup for GPU acceleration to work.

   4a) If installing globally, execute pip3 install --user pyyaml rospkg rospy std_msgs tensorflow matplotlib.

   4b) If installing locally, follow these steps as you see fit.

   1. Install the virtualenv package: sudo apt install python3-venv
   2. Create the virtualenv venv in the voice_recognition directory: virtualenv -p python3 venv
   3. Activate the virtualenv: source venv/bin/activate
   4. Install the required packages: pip install pyyaml rospkg rospy std_msgs tensorflow matplotlib.

   The virtualenv will automatically be activated when running bin/run.sh or bin/check_deps.sh. On the other hand, bin/volume_graph.py requires you to activate the virtualenv manually if matplotlib isn't installed globally.

Running

After running catkin_make, the package should be ready to launch with roslaunch voice_recognition main.launch. Another way to run the package is to just run bin/run.sh directly (cd to the bin/ folder first). (I think) this allows you to view more of the log output from the script. If you're just running the script directly, there is no need to call catkin_make. Just make sure that roscore is running.

If you're running this script on another machine than roscore, you'll need to manually edit ROS_MASTER_URI in bin/run.sh and point it to the IP/port of the roscore instance. I have not tried using roslaunch under this configuration, so I recommend manually editing the variable, and launching the script directly.

The script can also be run without ROS installed. If the packages rospy or std_msgs fail to import in Python, the script will output to the standard out stream. This behaviour can also be triggered by passing --stdout to run.sh, like so: ./run.sh --stdout.