EENAKWS (Robotics: R.2 Keyword spotting)
Our goal is to bring keyword spotting solutions to noisy environments, specializing on the on-site noise in an energy-efficient manner.
What is the challenge that is being addressed?
Formally, keyword spotting represents the process of recognizing predefined words from a speech signal. Our challenge was to perform such a task in noisy environments, such as factories or mines, allowing workers to control (heavy) machinery and reducing their exposure to loud, hazardous, and overall dangerous environments.
What is the AI solution the project plans to implement?
Our off-the-shelf EENAKWS AI solution is capable of recognizing keywords (e.g., stop, go, left, right) in noisy conditions, being generally robust to noises and being thus useful in controlling equipment or machinery in loud environments. Moreover, we allow for on-site specialization on the environmental noise; by doing this at runtime, our system can be continuously relocated, as it will keep adapting to the environment in which it is placed. Lastly, we ensured the mobility and extended lifetime of our AI solution by deploying it on (ultra)-low-power battery-operated devices, with each component of our application being energy efficient.
How will BonsAPPs support you in implementing this solution?
Throughout the past five months, BonsAPPs provided us with the tools to build, train, evaluate, and deploy our AI solution. Through a modular approach towards development, enabled by the BonsAPPs tools, we were able to train multiple neural networks on data from different dataset, including on data mixtures. Moreover, each of our so-obtained AI models could be deployed and benchmarked on different platforms, from computing clusters to ultra-low-power devices, thus obtaining AI solutions tailored to the users’ needs and requirements.