DigiEduBerlin team is happy to announce that first release of a small but powerful open-source automatic speech recognition (ASR) package "lesen-mikroserver" which is, from now on, available on Github here: https://github.com/hromi/lesen-mikroserver.

lesen-mikroserver is based on code from wonderful coqui speech-to-text (STT). server-side component of "lesen-mikroserver" is able to run inferences on 512-Volta tensor cores of Xavier Jetson Dev Kit GPUs devices and will be used exhaustively in the context of lesen.digital and fibel.digital projects.

Note also our small but tidy getUserMedia -> mediaRecorder -> Websocket -> Jetson -> Browser client which is more or less all You need to bring state of the art ASR / STT in Your pupil's browsers.