audio-attention

🤖 iCubAudioAttention 🔊

A collection of signal processing algorithms and audio capture tools implemented using YARP for the iCub humanoid robot.

Installation

• This repository depends on YARP (minimum version 3.2), icub-main, and icub-contrib-common, which can all be found on the robotology github. OpenCV is an optional dependency, for compiling visualisation tools. YCM (YARP CMake) is another optional dependency that offers extra CMake modules for YARP. If YCM is not found, local versions of the YCM modules will be used.
• For step-by-step installation instructions from a fresh install, you can follow our guide here.

git clone https://github.com/robotology/audio-attention.git
cd iCubAudioAttention && mkdir build && cd build
ccmake ..

• In the CCMake interface you can enable/disable the following flags:
  • BUILD_SHARED_LIBS → Compiles audioCubLib as a shared object, instead of static.
  • COMPILE_TESTS → Compiles some modules for testing other modules.
  • ENABLE_COMPILE_ON_PC104 → Only compiles modules that would be run on the robot.
  • ENABLE_OMP → Compiles processing modules with OpenMP multithreading.
  • ICUBCONTRIB_INSTALL_WITH_RPATH → Sets an rpath after installing modules.
  • ICUBCONTRIB_SHARED_LIBRARY → Compiles audioCubLib as a shared object, instead of static.

make -j
make install

Applications for yarpmanager will be installed in the ICUBcontrib directory in templates/applications. These should be modified and installed in the normal applications directory to be used.

Modules Description

Audio Preprocessor : Used to spectrally and spatially decompose a yarp sound object. Can optionally do additional processing to isolate for acoustic dynamics found in natural human speech.

Audio Bayesian Map : Takes output from the audioPreprocessor in the form of an allocentric map, and performs Bayesian updates by combining new evidence with a running prior knowledge of the auditory scene.

Audio Experiment : Contains a variety of tools and modules to streamline the process of recording audio and running experiments.

Frequency Visualisation : A module that receives a yarp matrix, and plots it as a yarp image. Allows live visualisation of the modules.

Remote Interface : A simple straight forward interface for streaming yarp sound objects recorded from the microphones on the robots head.