This github repo includes modelling of this pipeline along with other elements for more context from audio and reducing dependency on text as much as possible.