Spoken language understanding (SLU) topic has seen a lot of progress these\nlast three years, with the emergence of end-to-end neural approaches. Spoken\nlanguage understanding refers to natural language processing tasks related to\nsemantic extraction from speech signal, like named entity recognition from\nspeech or slot filling task in a context of human-machine dialogue.\nClassically, SLU tasks were processed through a cascade approach that consists\nin applying, firstly, an automatic speech recognition process, followed by a\nnatural language processing module applied to the automatic transcriptions.\nThese three last years, end-to-end neural approaches, based on deep neural\nnetworks, have been proposed in order to directly extract the semantics from\nspeech signal, by using a single neural model. More recent works on\nself-supervised training with unlabeled data open new perspectives in term of\nperformance for automatic speech recognition and natural language processing.\nIn this paper, we present a brief overview of the recent advances on the French\nMEDIA benchmark dataset for SLU, with or without the use of additional data. We\nalso present our last results that significantly outperform the current\nstate-of-the-art with a Concept Error Rate (CER) of 11.2%, instead of 13.6% for\nthe last state-of-the-art system presented this year.\n