Why shape is (1,n_values)? wont it be tx,n_values
That is just the input for one timestep. At each timestep you will have an input of that shape and an output of that shape. Those are the individual musical notes in this application. Then when you consider the entire sequence to get the sequence of notes which make up the tune and will have the shape you describe.
So basically the music_inference_model predicts sequence of notes based on input of shape (1,n_valuees) i.e on one note. then where is the dj_model used as we have trained that which has learned the patterns of a jazz soloist?
The dj_model is trained on full “songs” (sequences of notes). Then we use the pretrained model to implement music_inference_model based on inputting a single note and seeing what it predicts based on the pretrained coefficients learned from the songs.