Luxembourgish language corpus platform
A concordancer for the whole collection. It searches word forms, lemmas and normalised spellings, supports corpus query language for patterns over several tokens, and shows every hit in context with the recording playable at the hit. On the result set it computes statistics: frequency lists, n-grams, collocations, dispersion across texts and keyness against a reference corpus.
The management module. Curators build the corpus tree, upload recordings and texts, edit metadata, assign licences and permissions to roles, and export datasets, for example a speech corpus as audio segments with their transcripts, ready for training a recogniser.
A transcript editor for recordings. It plays the audio alongside the transcript, lets the transcriber correct words and sentence boundaries, and keeps the time alignment of every word to the sound. New recordings get a first transcript from the Sproochmaschinn speech recogniser, which the transcriber then corrects.
SproochHub is built on PostgreSQL, a Python API and a Vue.js web application. All text is indexed token by token in the database, and every query carries the licence and role rules that decide what each user may see.