Imparo IA

A Progressive Web App for teaching music with artificial intelligence in primary and lower secondary school, and in part in upper secondary school. The models run inside the browser on the classroom device, with no accounts and nothing leaving the room, and after the first load every station also works without a connection. It is the pupils' lab of the same line whose teacher course is Aula IA.

Six stations, each with a question of its own. The sound map has the class listen to one minute of the place they are in and mark every sound on a radial map, by distance and direction, with Schafer's questions and Krause's families. What the machine hears records ten seconds, shows the words the model recognises and how confident it is, and the pupils grade every word. Annotate in chunks has a soundscape annotated chunk by chunk and compares the class reading with the model's, with an export towards Soundscape Annotation Atelier. Continue the melody has an opening written note by note, with the note values and rests the class chooses, and the model carries it on with three degrees of freedom, all written on the staff and downloadable as MusicXML and MIDI. Where is the mistake puts two versions next to the real sound, only one of them manipulated, to train informed suspicion. Guess the instrument pits the ears of the class against the machine's on orchestra and band timbres.

The sound bank holds 87 clips, five recordings by the author and 82 CC0 samples, and the corpus of the mistake game is built in house, with a single declared manipulation per sound and the distance from the real sound measured. The notebook keeps what the class produces on the device, maps, sheets, scores and games, and exports it. The teacher page states what each station is for, what to observe and which questions to ask, and prepares the lesson for offline use by downloading the needed files in advance.

The texts come in two registers, one for primary and one for lower secondary school, pupils are addressed informally and the machine's ear speaks in the first person. The language-model tutor is reserved for the teacher and works with a personal key, everything else runs locally.