Update app.py

app.py

@@ -5,7 +5,6 @@ import datetime
 from pytz import timezone
 
 import torch
-import torch.nn.functional as F
 
 import gradio as gr
 import spaces
@@ -18,12 +17,11 @@ from midi_to_colab_audio import midi_to_colab_audio
 
 import matplotlib.pyplot as plt
 
-in_space = os.getenv("SYSTEM") == "spaces"
-
 # =================================================================================================
 
 @spaces.GPU
-def GenerateMIDI(num_tok, idrums, iinstr):
+def GenerateMIDI(num_tok, idrums, iinstr, input_align):
+    
     print('=' * 70)
     print('Req start time: {:%Y-%m-%d %H:%M:%S}'.format(datetime.datetime.now(PDT)))
     start_time = time.time()
@@ -32,6 +30,7 @@ def GenerateMIDI(num_tok, idrums, iinstr):
     print('Req num tok:', num_tok)
     print('Req instr:', iinstr)
     print('Drums:', idrums)
+    print('Align:', input_align)
     print('-' * 70)
 
     if idrums:
@@ -49,35 +48,7 @@ def GenerateMIDI(num_tok, idrums, iinstr):
     print(start_tokens)
     print('-' * 70)
 
-    output_signature = 'Allegro Music Transformer'
-    output_file_name = 'Allegro-Music-Transformer-Music-Composition'
-    track_name = 'Project Los Angeles'
-    list_of_MIDI_patches = [0, 24, 32, 40, 42, 46, 56, 71, 73, 0, 53, 19, 0, 0, 0, 0]
-    number_of_ticks_per_quarter = 500
-    text_encoding = 'ISO-8859-1'
-
-    output_header = [number_of_ticks_per_quarter,
-                     [['track_name', 0, bytes(output_signature, text_encoding)]]]
-
-    patch_list = [['patch_change', 0, 0, list_of_MIDI_patches[0]],
-                  ['patch_change', 0, 1, list_of_MIDI_patches[1]],
-                  ['patch_change', 0, 2, list_of_MIDI_patches[2]],
-                  ['patch_change', 0, 3, list_of_MIDI_patches[3]],
-                  ['patch_change', 0, 4, list_of_MIDI_patches[4]],
-                  ['patch_change', 0, 5, list_of_MIDI_patches[5]],
-                  ['patch_change', 0, 6, list_of_MIDI_patches[6]],
-                  ['patch_change', 0, 7, list_of_MIDI_patches[7]],
-                  ['patch_change', 0, 8, list_of_MIDI_patches[8]],
-                  ['patch_change', 0, 9, list_of_MIDI_patches[9]],
-                  ['patch_change', 0, 10, list_of_MIDI_patches[10]],
-                  ['patch_change', 0, 11, list_of_MIDI_patches[11]],
-                  ['patch_change', 0, 12, list_of_MIDI_patches[12]],
-                  ['patch_change', 0, 13, list_of_MIDI_patches[13]],
-                  ['patch_change', 0, 14, list_of_MIDI_patches[14]],
-                  ['patch_change', 0, 15, list_of_MIDI_patches[15]],
-                  ['track_name', 0, bytes(track_name, text_encoding)]]
-
-    output = output_header + [patch_list]
+    output = []
 
     print('Loading model...')
 
@@ -109,7 +80,7 @@ def GenerateMIDI(num_tok, idrums, iinstr):
 
     print('Done!')
     print('=' * 70)
-
+    print('Generating...')
 
     inp = torch.LongTensor([start_tokens]).cuda()
 
@@ -122,41 +93,51 @@ def GenerateMIDI(num_tok, idrums, iinstr):
                                         verbose=False)
 
     out0 = out[0].tolist()
+
+    patches = [0, 24, 32, 40, 42, 46, 56, 71, 73, 0, 53, 19, 0, 0, 0, 0]
     
     ctime = 0
-    dur = 0
+    dur = 1
     vel = 90
-    pitch = 0
+    pitch = 60
     channel = 0
 
     for ss1 in out0:
 
         if 0 < ss1 < 256:
+            
             ctime += ss1 * 8
     
         if 256 <= ss1 < 1280:
+            
             dur = ((ss1 - 256) // 8) * 32
             vel = (((ss1 - 256) % 8) + 1) * 15
     
         if 1280 <= ss1 < 2816:
+            
             channel = (ss1 - 1280) // 128
             pitch = (ss1 - 1280) % 128
             
             if channel != 9:
-                pat = list_of_MIDI_patches[channel]
+                pat = patches[channel]
             else:
                 pat = 128
                 
             event = ['note', ctime, dur, channel, pitch, vel, pat]
             
-            output[-1].append(event)
+            output.append(event)
 
-    midi_data = TMIDIX.score2midi(output, text_encoding)
+    if input_align:
+        output = TMIDIX.align_escore_notes_to_bars(output)
 
-    with open(f"Allegro-Music-Transformer-Composition.mid", 'wb') as f:
-        f.write(midi_data)
+    detailed_stats = TMIDIX.Tegridy_ms_SONG_to_MIDI_Converter(output,
+                                                              output_signature = 'Allegro Music Transformer',
+                                                              output_file_name = 'Allegro-Music-Transformer-Composition',
+                                                              track_name='Project Los Angeles',
+                                                              list_of_MIDI_patches=patches
+                                                              )
 
-    output_plot = TMIDIX.plot_ms_SONG(output[2], plot_title='Allegro-Music-Transformer-Composition', return_plt=True)
+    output_plot = TMIDIX.plot_ms_SONG(output, plot_title='Allegro-Music-Transformer-Composition', return_plt=True)
 
     audio = midi_to_colab_audio('Allegro-Music-Transformer-Composition.mid', 
                         soundfont_path="SGM-v2.01-YamahaGrand-Guit-Bass-v2.7.sf2",
@@ -184,8 +165,11 @@ if __name__ == "__main__":
     print('=' * 70)
     
     app = gr.Blocks()
+    
     with app:
+        
         gr.Markdown("<h1 style='text-align: center; margin-bottom: 1rem'>Allegro Music Transformer</h1>")
+        
         gr.Markdown(
             "![Visitors](https://api.visitorbadge.io/api/visitors?path=asigalov61.Allegro-Music-Transformer&style=flat)\n\n"
             "Full-attention multi-instrumental music transformer featuring asymmetrical encoding with octo-velocity, and chords counters tokens, optimized for speed and performance\n\n"
@@ -200,6 +184,7 @@ if __name__ == "__main__":
             ["Piano", "Guitar", "Bass", "Violin", "Cello", "Harp", "Trumpet", "Sax", "Flute", "Choir", "Organ"],
             value="Piano", label="Lead Instrument Controls", info="Desired lead instrument")
         input_drums = gr.Checkbox(label="Add Drums", value=False, info="Add drums to the composition")
+        input_align = gr.Checkbox(label="Align output to bars", value=False, info="Align output to bars")
         input_num_tokens = gr.Slider(16, 1024, value=512, label="Number of Tokens", info="Number of tokens to generate")
         
         run_btn = gr.Button("generate", variant="primary")
@@ -207,6 +192,7 @@ if __name__ == "__main__":
         output_audio = gr.Audio(label="output audio", format="mp3", elem_id="midi_audio")
         output_plot = gr.Plot(label='output plot')
         output_midi = gr.File(label="output midi", file_types=[".mid"])
-        run_event = run_btn.click(GenerateMIDI, [input_num_tokens, input_drums, input_instrument],
+        
+        run_event = run_btn.click(GenerateMIDI, [input_num_tokens, input_drums, input_instrument, input_align],
                                   [output_plot, output_midi, output_audio])
         app.queue().launch()