introlab/odas

ODAS: Open embeddeD Audition System

What is ODAS?

ODAS (Open embeddeD Audition System) is a free, open-source software library that gives machines a sense of hearing. It processes audio from microphones to figure out where sounds are coming from, follow those sounds as they move, separate different sounds from each other, and clean up the audio (post-filtering).

Think of it as the "ears" for a robot or smart device — it doesn't just record sound, it understands the spatial scene.

Key facts

Field Detail
What it does Sound source localization (where is the sound?), tracking (how does it move?), separation (which sound is which?), and post-filtering (cleaning up the signal).
Language Written entirely in C, chosen for portability across different hardware.
Target hardware Optimized to run on low-cost embedded hardware — small, cheap computers inside devices.
License MIT license — permissive open source, free to use and modify.
Documentation A wiki explains how to build and run the software.
Demo A YouTube video shows ODAS in action.

Related projects (from the README)

  • odas_ros — a separate project that integrates ODAS with ROS (Robot Operating System), a common framework for robotics.
  • odas_web — a graphical interface for visualizing the live data ODAS produces (e.g., showing where sounds are detected).
  • 8SoundsUSB — open-source hardware: a USB-powered microphone array with 8 inputs, configurable.
  • 16SoundsUSB — a bigger version with 16 inputs, also USB-powered and configurable.

How it works (in plain terms)

ODAS takes audio from multiple microphones arranged in an array. By comparing the tiny timing and volume differences between microphones, it can:

  1. Locate — determine the direction and distance of a sound source.
  2. Track — keep following that source as it moves.
  3. Separate — isolate individual sound sources from a mixture (e.g., one voice from background noise).
  4. Post-filter — reduce noise and improve the quality of the separated audio.

Academic background

Two papers describe the methods used in ODAS:

  • 2022 — "ODAS: Open embeddeD Audition System" (published in Frontiers in Robotics and AI).
  • 2019 — "Lightweight and Optimized Sound Source Localization and Tracking Methods for Opened and Closed Microphone Array Configurations" (published in Robotics and Autonomous Systems).

The 2019 paper highlights that ODAS is designed to be lightweight and optimized — meaning it doesn't waste computing power, which is why it can run on cheap hardware.

Bottom line

ODAS is a compact, efficient, open-source "hearing system" for robots and embedded devices. It turns raw microphone input into useful spatial audio information — where sounds are, how they move, and how to separate them — all in a package small enough to run on modest hardware.

Related

  • Project
  • Project
  • Project
  • Project