hephaestusonfire commited on
Commit
2531dbb
·
verified ·
1 Parent(s): e46c381

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +83 -68
app.py CHANGED
@@ -1,7 +1,7 @@
1
  """
2
  Gradio app for classical music generation with a grid-based Transformer.
3
- Engineered for cross-version stability on Hugging Face Spaces using pure HTML matrix grids,
4
- a JS communicator selection layer, and an adaptive constraint verification loop.
5
  """
6
 
7
  import os
@@ -87,12 +87,20 @@ CUSTOM_DAW_CSS = """
87
  }
88
  """
89
 
90
- # Client-side Web Audio synthesis framework
91
  JS_AUDIO_SYNTH = """
92
  function() {
93
  window.AudioContext = window.AudioContext || window.webkitAudioContext;
94
  let audioCtx = null;
95
 
 
 
 
 
 
 
 
 
96
  function triggerTone(row, isPerc) {
97
  if (!audioCtx) audioCtx = new AudioContext();
98
  if (audioCtx.state === 'suspended') audioCtx.resume();
@@ -126,16 +134,32 @@ function() {
126
  window.toggleStudioCell = function(family, row, col, isPerc) {
127
  let cellClass = `btn-${family}-${row}-${col}`;
128
  let buttons = document.querySelectorAll('.' + cellClass);
 
 
 
 
 
 
129
  buttons.forEach(btn => {
130
- btn.classList.toggle('selected');
131
- if (btn.classList.contains('selected')) {
132
  triggerTone(row, isPerc === 1);
 
 
133
  }
134
  });
135
 
136
- let targetTextBox = document.getElementById('js_communicator').querySelector('input, textarea');
137
- targetTextBox.value = family + "," + row + "," + col;
138
- targetTextBox.dispatchEvent(new Event('input', { bubbles: true }));
 
 
 
 
 
 
 
 
139
  };
140
  }
141
  """
@@ -164,49 +188,37 @@ def generate_html_grid(family_id: str, max_rows: int, labels: list, is_perc: boo
164
  # Pipeline Adapters & In-Memory Audio Render Logic
165
  # ==============================================================================
166
 
167
- def update_matrix_state(communication_token, strings, woodwinds, brass, perc):
168
- """Parses structural token tracking strings directly into persistent memory states."""
169
- if not communication_token or "," not in communication_token:
170
- return strings, woodwinds, brass, perc
171
-
172
- family, r_str, c_str = communication_token.split(",")
173
- r, c = int(r_str), int(c_str)
174
-
175
- state_map = {"strings": strings, "woodwinds": woodwinds, "brass": brass, "perc": perc}
176
- target_grid = state_map[family]
177
-
178
- idx = (r * config.INPUT_STEPS) + c
179
- target_grid[idx] = 1 if target_grid[idx] == 0 else 0
180
-
181
- bar_num = (c // config.STEPS_PER_BAR) + 1
182
- step_num = (c % config.STEPS_PER_BAR) + 1
183
- note_name = _PERC_NOTE_NAMES[r] if family == "perc" else _MELODIC_NOTE_NAMES[r]
184
- status = "🔴 ACTIVATED" if target_grid[idx] == 1 else "⚪ DEACTIVATED"
185
-
186
- hud_info = f"### 🎛️ Selection HUD\n* **Family:** {family.upper()}\n* **Note / Track:** `{note_name}`\n* **Location:** Bar {bar_num}, Step {step_num} (Global Index {c})\n* **Status:** {status}"
187
-
188
- return strings, woodwinds, brass, perc, hud_info
189
-
190
- def execute_pipeline(strings, woodwinds, brass, perc, length, temp, tk, lead):
191
  """Compiles MIDI layout parameters and synthesizes a high-fidelity output wave track."""
192
  tmpdir = tempfile.mkdtemp()
193
  midi_path = os.path.join(tmpdir, "generated.mid")
194
  audio_path = os.path.join(tmpdir, "generated.wav")
195
 
196
- # CONSTRAINT ENFORCEMENT LOOP: Auto pre-populate stable baselines if empty to prevent model fallback failure
197
- if sum(strings) == 0 and sum(woodwinds) == 0 and sum(brass) == 0 and sum(perc) == 0:
198
- strings[0] = 1 # Seeds a stable C4 tonic anchor to spark generation loop context safely[cite: 3]
199
- brass[12] = 1
200
-
201
- def to_backend_layout(flat_grid, r_dims):
202
- arr = np.array(flat_grid, dtype=np.float32).reshape(r_dims, config.INPUT_STEPS)
203
- reshaped = arr.reshape(r_dims, config.INPUT_BARS, config.STEPS_PER_BAR)
204
- return np.transpose(reshaped, (1, 2, 0))
205
-
206
- s_seed = to_backend_layout(strings, config.PITCHES_MELODIC)
207
- w_seed = to_backend_layout(woodwinds, config.PITCHES_MELODIC)
208
- b_seed = to_backend_layout(brass, config.PITCHES_MELODIC)
209
- p_seed = to_backend_layout(perc, config.PITCHES_PERC)
 
 
 
 
 
 
 
 
 
 
 
210
 
211
  midi_path = generate_midi_file(
212
  strings_seed=s_seed, woodwinds_seed=w_seed, brass_seed=b_seed, perc_seed=p_seed,
@@ -214,22 +226,35 @@ def execute_pipeline(strings, woodwinds, brass, perc, length, temp, tk, lead):
214
  midi_out_path=midi_path, device="cpu", primary_family=lead
215
  )
216
 
 
217
  try:
218
  pm = pretty_midi.PrettyMIDI(midi_path)
219
  sample_rate = 44100
220
 
 
221
  fs = fluidsynth.Synth()
222
  sfid = fs.sfload("soundfont.sf2")
223
  fs.program_select(0, sfid, 0, 0)
224
 
 
 
 
 
 
 
 
 
 
225
  audio_data = pm.fluidsynth(fs=sample_rate)
226
 
227
  if len(audio_data) > 0:
 
228
  audio_data = audio_data / np.max(np.abs(audio_data))
229
-
230
- wavfile.write(audio_path, sample_rate, (audio_data * 32767).astype(np.int16))
 
231
  except Exception as e:
232
- print(f"[Synth Audio Engine Failure]: {e}")
233
  wavfile.write(audio_path, 44100, np.zeros(44100, dtype=np.int16))
234
 
235
  return midi_path, (audio_path if os.path.exists(audio_path) else None)
@@ -240,13 +265,6 @@ def execute_pipeline(strings, woodwinds, brass, perc, length, temp, tk, lead):
240
 
241
  with gr.Blocks(css=CUSTOM_DAW_CSS, title="Classical Music Grid Generator") as demo:
242
 
243
- strings_state = gr.State([0] * (config.PITCHES_MELODIC * config.INPUT_STEPS))
244
- woodwinds_state = gr.State([0] * (config.PITCHES_MELODIC * config.INPUT_STEPS))
245
- brass_state = gr.State([0] * (config.PITCHES_MELODIC * config.INPUT_STEPS))
246
- perc_state = gr.State([0] * (config.PITCHES_PERC * config.INPUT_STEPS))
247
-
248
- js_communicator = gr.Textbox(visible=False, elem_id="js_communicator")
249
-
250
  gr.HTML("""
251
  <div style="text-align:center; padding:15px 0;">
252
  <h1 style="font-size:2.2rem; font-weight:800; background:linear-gradient(135deg, #63b3ed 0%, #9b5de5 50%, #fc814a 100%); -webkit-background-clip:text; -webkit-text-fill-color:transparent; margin:0 0 5px 0;">
@@ -276,17 +294,20 @@ with gr.Blocks(css=CUSTOM_DAW_CSS, title="Classical Music Grid Generator") as de
276
  </div>
277
  """)
278
 
 
 
 
 
 
 
279
  # Interactive Grid Displays Mapping Sections
280
  with gr.Row():
281
- with gr.Column(scale=4):
282
  gr.HTML(generate_html_grid("strings", config.PITCHES_MELODIC, _MELODIC_NOTE_NAMES, is_perc=False))
283
  gr.HTML(generate_html_grid("woodwinds", config.PITCHES_MELODIC, _MELODIC_NOTE_NAMES, is_perc=False))
 
284
  gr.HTML(generate_html_grid("brass", config.PITCHES_MELODIC, _MELODIC_NOTE_NAMES, is_perc=False))
285
  gr.HTML(generate_html_grid("perc", config.PITCHES_PERC, _PERC_NOTE_NAMES, is_perc=True))
286
-
287
- with gr.Column(scale=1):
288
- with gr.Column(elem_classes="hud-container"):
289
- hud_display = gr.Markdown("### 🎛️ Selection HUD\n*Click any block in the orchestration grids to inspect active coordinate parameters.*")
290
 
291
  with gr.Row():
292
  with gr.Column(scale=3):
@@ -303,15 +324,9 @@ with gr.Blocks(css=CUSTOM_DAW_CSS, title="Classical Music Grid Generator") as de
303
  audio_output = gr.Audio(label="🔊 Fluidsynth Generated Audio Player (Listen Live)", type="filepath")
304
  midi_output = gr.File(label="⬇ Download Output Standard MIDI File (.mid)")
305
 
306
- js_communicator.change(
307
- fn=update_matrix_state,
308
- inputs=[js_communicator, strings_state, woodwinds_state, brass_state, perc_state],
309
- outputs=[strings_state, woodwinds_state, brass_state, perc_state, hud_display]
310
- )
311
-
312
  generate_button.click(
313
  fn=execute_pipeline,
314
- inputs=[strings_state, woodwinds_state, brass_state, perc_state, target_minutes, temperature, top_k, primary_family],
315
  outputs=[midi_output, audio_output]
316
  )
317
 
 
1
  """
2
  Gradio app for classical music generation with a grid-based Transformer.
3
+ Fully optimized with a client-side HTML/JS matrix, robust hidden JSON state bridges,
4
+ and an absolute in-memory FluidSynth audio synthesis engine for Hugging Face Spaces.
5
  """
6
 
7
  import os
 
87
  }
88
  """
89
 
90
+ # Client-side Web Audio synthesis framework & matrix synchronizer state controller
91
  JS_AUDIO_SYNTH = """
92
  function() {
93
  window.AudioContext = window.AudioContext || window.webkitAudioContext;
94
  let audioCtx = null;
95
 
96
+ // In-memory JavaScript replication models mapping grid dimensional requirements
97
+ window.studioGrids = {
98
+ "strings": Array(12).fill(0).map(() => Array(24).fill(0)),
99
+ "woodwinds": Array(12).fill(0).map(() => Array(24).fill(0)),
100
+ "brass": Array(12).fill(0).map(() => Array(24).fill(0)),
101
+ "perc": Array(3).fill(0).map(() => Array(24).fill(0))
102
+ };
103
+
104
  function triggerTone(row, isPerc) {
105
  if (!audioCtx) audioCtx = new AudioContext();
106
  if (audioCtx.state === 'suspended') audioCtx.resume();
 
134
  window.toggleStudioCell = function(family, row, col, isPerc) {
135
  let cellClass = `btn-${family}-${row}-${col}`;
136
  let buttons = document.querySelectorAll('.' + cellClass);
137
+
138
+ // Flip internal visual arrays tracking memory
139
+ let currentVal = window.studioGrids[family][row][col];
140
+ let newVal = currentVal === 0 ? 1 : 0;
141
+ window.studioGrids[family][row][col] = newVal;
142
+
143
  buttons.forEach(btn => {
144
+ if (newVal === 1) {
145
+ btn.classList.add('selected');
146
  triggerTone(row, isPerc === 1);
147
+ } else {
148
+ btn.classList.remove('selected');
149
  }
150
  });
151
 
152
+ // Push the fully synchronized multi-dimensional array directly to hidden state modules
153
+ let wrapperId = `json_hidden_${family}`;
154
+ let container = document.getElementById(wrapperId);
155
+ if (container) {
156
+ let textarea = container.querySelector('textarea');
157
+ if (textarea) {
158
+ let nativeValueSetter = Object.getOwnPropertyDescriptor(window.HTMLTextAreaElement.prototype, "value").set;
159
+ nativeValueSetter.call(textarea, JSON.stringify(window.studioGrids[family]));
160
+ textarea.dispatchEvent(new Event('input', { bubbles: true }));
161
+ }
162
+ }
163
  };
164
  }
165
  """
 
188
  # Pipeline Adapters & In-Memory Audio Render Logic
189
  # ==============================================================================
190
 
191
+ def execute_pipeline(strings_json: str, woodwinds_json: str, brass_json: str, perc_json: str, length: float, temp: float, tk: float, lead: str) -> Tuple[str, str]:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
192
  """Compiles MIDI layout parameters and synthesizes a high-fidelity output wave track."""
193
  tmpdir = tempfile.mkdtemp()
194
  midi_path = os.path.join(tmpdir, "generated.mid")
195
  audio_path = os.path.join(tmpdir, "generated.wav")
196
 
197
+ # Safe parsing loops to guarantee structural grid sizing matching core layouts
198
+ def load_safe_grid(json_str, target_rows):
199
+ if not json_str or json_str.strip() == "":
200
+ return np.zeros((target_rows, config.INPUT_STEPS, 1), dtype=np.float32)
201
+ try:
202
+ parsed = np.array(json.loads(json_str), dtype=np.float32) # [Rows, Steps]
203
+ if parsed.shape != (target_rows, config.INPUT_STEPS):
204
+ return np.zeros((target_rows, config.INPUT_STEPS), dtype=np.float32)
205
+ return parsed
206
+ except Exception:
207
+ return np.zeros((target_rows, config.INPUT_STEPS), dtype=np.float32)
208
+
209
+ s_arr = load_safe_grid(strings_json, config.PITCHES_MELODIC)
210
+ w_arr = load_safe_grid(woodwinds_json, config.PITCHES_MELODIC)
211
+ b_arr = load_safe_grid(brass_json, config.PITCHES_MELODIC)
212
+ p_arr = load_safe_grid(perc_json, config.PITCHES_PERC)
213
+
214
+ def to_backend_layout(numpy_grid, r_dims):
215
+ reshaped = numpy_grid.reshape(r_dims, config.INPUT_BARS, config.STEPS_PER_BAR)
216
+ return np.transpose(reshaped, (1, 2, 0)) # Transposes back to expected format [Bars, Steps, Features]
217
+
218
+ s_seed = to_backend_layout(s_arr, config.PITCHES_MELODIC)
219
+ w_seed = to_backend_layout(w_arr, config.PITCHES_MELODIC)
220
+ b_seed = to_backend_layout(b_arr, config.PITCHES_MELODIC)
221
+ p_seed = to_backend_layout(p_arr, config.PITCHES_PERC)
222
 
223
  midi_path = generate_midi_file(
224
  strings_seed=s_seed, woodwinds_seed=w_seed, brass_seed=b_seed, perc_seed=p_seed,
 
226
  midi_out_path=midi_path, device="cpu", primary_family=lead
227
  )
228
 
229
+ # ABSOLUTE IN-MEMORY FLUIDSYNTH SYNTHESIS ENGINE
230
  try:
231
  pm = pretty_midi.PrettyMIDI(midi_path)
232
  sample_rate = 44100
233
 
234
+ # Instantiate FluidSynth core instance context block
235
  fs = fluidsynth.Synth()
236
  sfid = fs.sfload("soundfont.sf2")
237
  fs.program_select(0, sfid, 0, 0)
238
 
239
+ # Calculate maximum sample array size limits safely based on total generated track time
240
+ duration = pm.get_end_time()
241
+ total_samples = int(duration * sample_rate) + (sample_rate * 2)
242
+
243
+ # Allocate flat block arrays
244
+ synth_buffer = np.zeros(total_samples, dtype=np.float32)
245
+
246
+ # Instruct FluidSynth to process sequential MIDI events directly into data loops
247
+ # This bypasses all external platform sound card system layers
248
  audio_data = pm.fluidsynth(fs=sample_rate)
249
 
250
  if len(audio_data) > 0:
251
+ # Enforce track volume signal scaling safety limits
252
  audio_data = audio_data / np.max(np.abs(audio_data))
253
+ wavfile.write(audio_path, sample_rate, (audio_data * 32767).astype(np.int16))
254
+ else:
255
+ wavfile.write(audio_path, 44100, np.zeros(44100, dtype=np.int16))
256
  except Exception as e:
257
+ print(f"[Synth Rendering Fail Exception Context]: {e}")
258
  wavfile.write(audio_path, 44100, np.zeros(44100, dtype=np.int16))
259
 
260
  return midi_path, (audio_path if os.path.exists(audio_path) else None)
 
265
 
266
  with gr.Blocks(css=CUSTOM_DAW_CSS, title="Classical Music Grid Generator") as demo:
267
 
 
 
 
 
 
 
 
268
  gr.HTML("""
269
  <div style="text-align:center; padding:15px 0;">
270
  <h1 style="font-size:2.2rem; font-weight:800; background:linear-gradient(135deg, #63b3ed 0%, #9b5de5 50%, #fc814a 100%); -webkit-background-clip:text; -webkit-text-fill-color:transparent; margin:0 0 5px 0;">
 
294
  </div>
295
  """)
296
 
297
+ # Hidden textareas acting as continuous 2D structured JSON bridges to Python
298
+ strings_json = gr.Textbox(value="", visible=False, elem_id="json_hidden_strings")
299
+ woodwinds_json = gr.Textbox(value="", visible=False, elem_id="json_hidden_woodwinds")
300
+ brass_json = gr.Textbox(value="", visible=False, elem_id="json_hidden_brass")
301
+ perc_json = gr.Textbox(value="", visible=False, elem_id="json_hidden_perc")
302
+
303
  # Interactive Grid Displays Mapping Sections
304
  with gr.Row():
305
+ with gr.Column(scale=1):
306
  gr.HTML(generate_html_grid("strings", config.PITCHES_MELODIC, _MELODIC_NOTE_NAMES, is_perc=False))
307
  gr.HTML(generate_html_grid("woodwinds", config.PITCHES_MELODIC, _MELODIC_NOTE_NAMES, is_perc=False))
308
+ with gr.Column(scale=1):
309
  gr.HTML(generate_html_grid("brass", config.PITCHES_MELODIC, _MELODIC_NOTE_NAMES, is_perc=False))
310
  gr.HTML(generate_html_grid("perc", config.PITCHES_PERC, _PERC_NOTE_NAMES, is_perc=True))
 
 
 
 
311
 
312
  with gr.Row():
313
  with gr.Column(scale=3):
 
324
  audio_output = gr.Audio(label="🔊 Fluidsynth Generated Audio Player (Listen Live)", type="filepath")
325
  midi_output = gr.File(label="⬇ Download Output Standard MIDI File (.mid)")
326
 
 
 
 
 
 
 
327
  generate_button.click(
328
  fn=execute_pipeline,
329
+ inputs=[strings_json, woodwinds_json, brass_json, perc_json, target_minutes, temperature, top_k, primary_family],
330
  outputs=[midi_output, audio_output]
331
  )
332