README.md

README.md

@@ -12,5 +12,16 @@ license: mit
 This project implements a GPT-2 based model for generating MIDI music sequences
 using deep learning. The model is trained on MIDI files from the MAESTRO
 dataset, converted to a special JSON format. The model uses HuggingFace's
-transformers library for training.
-Select a composer from the drop down and let the model create a unique musical sequence in their style.
+transformers library for training. 
+
+Select a composer from the dropdown and click generate music. This will initiate the 
+generation. and produce a short musical clip which can be downloaded. If the song is 
+to your liking, click continue sequence, which will add about 10 seconds to the song. 
+This continue process can be repeated to extend the song as much as you'd like. If the 
+initial sequence isn't inspiring, click generate music to start again and create a new 
+initial sequence.
+
+Some post processing on the generated sequence is done to ensure a 
+properly formatted midi file, however the model occasionally produces a file  that
+cannot be interpretted for playback. If this happens, simply begin again by clicking
+generate music.