What exactly is AMSI?
In Zoundream, we've spent several months to determine the best solution to identify and translate a human produced sound into its actual meaning, specifically for children pre-speech. There are several approaches in Machine Learning literature to address this challenge, but unfortunately none of them is actually suitable to work on a continuous incoming stream of audio. So we created our custom deep-learning based solution. AMSI, which is an acronym for Acoustic MultiStage Interpreter, processes the sound into a series of inner deep-learning models to extract the meaning of a baby cry. Think about understanding letters, then words, then meaningful sentences. That's exactly what AMSI does. AMSI can be adapted to many different problems and shortly we'll be able to indentify pathologies and uncommon behaviours of the baby.
Adapting Our Technology in Your Product
We successfully prototyped AMSI on a little hardware device that we made for testing purposes only. This constraint has forced us to optimize its engine for many potential architectures such different embedded hardware platforms or mobile systems. But we are currently porting AMSI to a remote service-based solution to take advantage of the increased computational power of Cloud Computing and provide even more accurate results.