Upload app.py

app.py

@@ -129,44 +129,53 @@ def generate_midi(temperature, top_k):
 
     midi_events = round_bars
 
+    # remove duplciates
+    for track in midi_events:
+        track.sort(key=lambda x: x['start'])
+        unique_notes = []
+        
+        for note in track:
+            if not any(abs(note['start'] - n['start']) < 12 and note['pitch'] == n['pitch'] for n in unique_notes):
+                unique_notes.append(note)
+        
+        track[:] = unique_notes 
+
     return midi_events
 
 
-def write_midi(midi_events, bpm):
+def write_single_midi(midi_events, bpm):
     midi_data = pretty_midi.PrettyMIDI(initial_tempo=bpm, resolution=96)
     midi_data.time_signature_changes.append(pretty_midi.containers.TimeSignature(4, 4, 0))
     instrument = pretty_midi.Instrument(0)
     midi_data.instruments.append(instrument)
 
-    for sequence in midi_events:
-        for event in sequence:
-            pitch = event['pitch']
-            velocity = event['velocity']
-            start = midi_data.tick_to_time(event['start'])
-            end = midi_data.tick_to_time(event['end'])
-            note = pretty_midi.Note(pitch=pitch, velocity=velocity, start=start, end=end)
-            instrument.notes.append(note)
+    for event in midi_events[0]:
+        pitch = event['pitch']
+        velocity = event['velocity']
+        start = midi_data.tick_to_time(event['start'])
+        end = midi_data.tick_to_time(event['end'])
+        note = pretty_midi.Note(pitch=pitch, velocity=velocity, start=start, end=end)
+        instrument.notes.append(note)
 
     midi_path = os.path.join(temp_dir, 'output.mid')
     midi_data.write(midi_path)
     print(f"Generated: {midi_path}")
 
 
-def render_wav(midi_file, uploaded_sf2=None):
+def render_wav(midi_file, uploaded_sf2=None, output_level='2.0'):
     sf2_dir = 'sf2_kits'
     audio_format = 's16'
     sample_rate = '44100'
-    gain = '2.0'
+    gain = str(output_level)
 
     if uploaded_sf2:
         sf2_file = uploaded_sf2
     else:
-        sf2_files = [f for f in os.listdir(sf2_dir) if f.endswith('.sf2')]
+        sf2_files = [f for f in os.listdir(os.path.join(sf2_dir)) if f.endswith('.sf2')]
         if not sf2_files:
             raise ValueError("No SoundFont (.sf2) file found in directory.")
         sf2_file = os.path.join(sf2_dir, random.choice(sf2_files))
 
-    print(f"Using SoundFont: {sf2_file}")
     output_wav = os.path.join(temp_dir, 'output.wav')
 
     with open(os.devnull, 'w') as devnull:
@@ -179,15 +188,15 @@ def render_wav(midi_file, uploaded_sf2=None):
     return output_wav
 
 
-def generate_and_return_files(bpm, temperature, top_k, uploaded_sf2=None):
+def generate_and_return_files(bpm, temperature, top_k, uploaded_sf2=None, output_level='2.0'):
     midi_events = generate_midi(temperature, top_k)  
     if not midi_events:
         return "Error generating MIDI.", None, None
     
-    write_midi(midi_events, bpm)
+    write_single_midi(midi_events, bpm)
     
     midi_file = os.path.join(temp_dir, 'output.mid')
-    wav_raw = render_wav(midi_file, uploaded_sf2)
+    wav_raw = render_wav(midi_file, uploaded_sf2, output_level)
     wav_fx = os.path.join(temp_dir, 'output_fx.wav')
 
     sfx_settings = [
@@ -213,40 +222,60 @@ def generate_and_return_files(bpm, temperature, top_k, uploaded_sf2=None):
 
 
 custom_css = """
+.gradio-container {
+  max-width: 1200px !important;
+  margin: 0 auto;
+}
+
+input[type="range"] {
+  background-image: linear-gradient(#7c4dff, #7c4dff);
+}
+
 #generate-btn {
-    background-color: #6366f1 !important;
-    color: white !important;
-    border: none !important;
-    font-size: 16px;
-    padding: 10px 20px;
-    border-radius: 5px;
-    cursor: pointer;
+  font-size: 18px;
+  padding: 10px 20px;
+  border-radius: 5px;
+  cursor: pointer;
+  background: linear-gradient(90deg, hsla(268, 90%, 68%, 1) 0%, hsla(260, 72%, 70%, 1) 50%, hsla(247, 73%, 65%, 1) 100%);
+  transition: background 1s ease;
 }
+
 #generate-btn:hover {
-    background-color: #4f51c5 !important;
+  background: linear-gradient(90deg, hsla(268, 90%, 78%, 1) 0%, hsla(260, 72%, 80%, 1) 50%, hsla(247, 73%, 75%, 1) 100%);
+}
+
+#component-11 .download a {
+  font-size: 16px;
 }
 """
 
-with gr.Blocks(css=custom_css, theme="soft") as iface:
-    gr.Markdown("<h1 style='font-weight: bold; text-align: center;'>nanoMPC - AI Midi Drum Sequencer</h1>")
-    gr.Markdown("<p style='text-align:center;'>nanoMPC is a tiny transformer model that generates MIDI drum beats inspired by Lo-Fi, Boom Bap and other styles of Hip Hop.</p>")
+waveform_opts = gr.WaveformOptions(
+    waveform_color="#888888",
+    waveform_progress_color="#7c4dff",
+    trim_region_color="#7c4dff"
+)
+
+with gr.Blocks(css=custom_css, theme=gr.themes.Monochrome(font=gr.themes.GoogleFont("Roboto"))) as iface:
+    gr.Markdown("<h1 style='font-weight: bold; text-align: center; font-size: 40px; margin: 0px;'>nanoMPC</h1>")
+    gr.Markdown("<p style='text-align: center; font-size: 18px;'>nanoMPC is a MIDI transformer model that generates lo-fi and boom bap beats.</p>")
     
     with gr.Row():
         with gr.Column(scale=1):
-            bpm = gr.Slider(minimum=50, maximum=200, step=1, value=90, label="BPM")
+            bpm = gr.Slider(minimum=50, maximum=200, step=1, value=100, label="BPM")
             temperature = gr.Slider(minimum=0.1, maximum=2.0, step=0.1, value=1.0, label="Temperature")
-            top_k = gr.Slider(minimum=4, maximum=128, step=1, value=64, label="Top-k")
-            soundfont = gr.File(label="Optional: Upload SoundFont (preset=0, bank=0)")
-        
-        with gr.Column(scale=1):
-            midi_file = gr.File(label="MIDI File Output")
-            audio_file = gr.Audio(label="Generated Audio Output", type="filepath")
+            top_k = gr.Slider(minimum=4, maximum=16, step=1, value=8, label="Top-k")
+            output_level = gr.Slider(minimum=0, maximum=3, step=0.10, value=2.0, label="Output Gain")
+            midi_file = gr.File(label="MIDI Output")
+            audio_file = gr.Audio(label="Audio Output", type="filepath", waveform_options=waveform_opts)
             generate_button = gr.Button("Generate", elem_id="generate-btn")
+            soundfont = gr.File(label="Optional: Upload SoundFont (preset=0, bank=0)")
     
     generate_button.click(
         fn=generate_and_return_files,
-        inputs=[bpm, temperature, top_k, soundfont],
+        inputs=[bpm, temperature, top_k, soundfont, output_level],
         outputs=[midi_file, audio_file]
     )
 
+    gr.Markdown("<p style='text-align: center; font-size: 14px;'>Developed by <a href='https://www.patchbanks.com/' target='_blank'><strong>Patchbanks</strong></a></p>")
+
 iface.launch(share=True)