Commit
·
049cfd4
1
Parent(s):
836c70c
moved to github.io
Browse files- app.py +27 -333
- generation/__pycache__/gen_utils.cpython-39.pyc +0 -0
- model/__pycache__/__init__.cpython-39.pyc +0 -0
- model/__pycache__/latent_diffusion.cpython-39.pyc +0 -0
- model/__pycache__/model_sdf.cpython-39.pyc +0 -0
- model/__pycache__/sampler_sdf.cpython-39.pyc +0 -0
- model/architecture/__pycache__/unet.cpython-39.pyc +0 -0
- model/architecture/__pycache__/unet_attention.cpython-39.pyc +0 -0
- output_0.mid +0 -0
- output_0.wav +2 -2
- piano_roll.png +2 -2
- train/__pycache__/__init__.cpython-39.pyc +0 -0
- train/__pycache__/learner.cpython-39.pyc +0 -0
- train/__pycache__/train_params.cpython-39.pyc +0 -0
app.py
CHANGED
@@ -57,10 +57,8 @@ def update_musescore_image(selected_prompt):
|
|
57 |
elif selected_prompt == "example 6":
|
58 |
return "samples/diy_examples/example6/example6.jpg"
|
59 |
|
60 |
-
|
61 |
-
# Model for generating music (example)
|
62 |
def generate_music(prompt, tempo, num_samples=1, mode="example", rhythm_control="Yes"):
|
63 |
-
|
64 |
ldm_model = init_ldm_model(params_chord_lsh_cond, debug_mode=False)
|
65 |
model = Diffpro_SDF.load_trained(ldm_model, model_path).to(device)
|
66 |
sampler = SDFSampler(model.ldm, 64, 64, is_autocast=False, device=device, debug_mode=False)
|
@@ -101,7 +99,6 @@ def generate_music(prompt, tempo, num_samples=1, mode="example", rhythm_control=
|
|
101 |
if background_condition[:,4:6,:,:].min()>=0:
|
102 |
full_lsh_roll = extend_piano_roll(background_condition[i,4:6,:,:].cpu().numpy())
|
103 |
midi_file = piano_roll_to_midi(full_roll, full_chd_roll, full_lsh_roll, bpm=tempo)
|
104 |
-
# filename = f'DDIM_w_rhythm_onset_0to10_{i}_edit_x0_and_eps'+'.mid'
|
105 |
filename = f"output_{i}.mid"
|
106 |
save_midi(midi_file, filename)
|
107 |
subprocess.Popen(['timidity',f'output_{i}.mid','-Ow','-o',f'output_{i}.wav']).communicate()
|
@@ -126,129 +123,16 @@ def visualize_midi(midi):
|
|
126 |
plt.savefig(output_image_path)
|
127 |
return output_image_path
|
128 |
|
129 |
-
def plot_rhythm(rhythm_str, label):
|
130 |
-
if rhythm_str=="null rhythm":
|
131 |
-
return None
|
132 |
-
fig, ax = plt.subplots(figsize=(6, 2))
|
133 |
-
|
134 |
-
# Ensure it's a 16-bit string
|
135 |
-
rhythm_str = rhythm_str[:16]
|
136 |
-
|
137 |
-
# Convert string to a list of 0s and 1s
|
138 |
-
rhythm = [0 if bit=="0" else 1 for bit in rhythm_str]
|
139 |
-
|
140 |
-
# Define the x axis for the 16 sixteenth notes
|
141 |
-
x = list(range(1, 17)) # 1 to 16 sixteenth notes
|
142 |
-
|
143 |
-
# Plot each note (1 as filled circle, 0 as empty circle)
|
144 |
-
for i, bit in enumerate(rhythm):
|
145 |
-
if bit == 1:
|
146 |
-
ax.scatter(i + 1, 1, color='black', s=100, label="Note" if i == 0 else "")
|
147 |
-
else:
|
148 |
-
ax.scatter(i + 1, 1, edgecolor='black', facecolor='none', s=100, label="Rest" if i == 0 else "")
|
149 |
-
|
150 |
-
# Distinguish groups of 4 using vertical dashed lines (no solid grid lines)
|
151 |
-
for i in range(4, 17, 4):
|
152 |
-
ax.axvline(x=i + 0.5, color='grey', linestyle='--')
|
153 |
-
|
154 |
-
# Remove solid vertical grid lines by setting the grid off
|
155 |
-
ax.grid(False)
|
156 |
-
|
157 |
-
# Formatting the plot
|
158 |
-
ax.set_xlim(0.5, 16.5)
|
159 |
-
ax.set_ylim(0.8, 1.2)
|
160 |
-
ax.set_xticks(x)
|
161 |
-
ax.set_yticks([])
|
162 |
-
ax.set_xlabel("16th Notes")
|
163 |
-
ax.set_title("Rhythm Pattern")
|
164 |
-
|
165 |
-
fig.savefig(f'samples/diy_examples/rhythm_plot_{label}.png')
|
166 |
-
plt.close(fig)
|
167 |
-
return f'samples/diy_examples/rhythm_plot_{label}.png'
|
168 |
-
|
169 |
-
def adjust_rhythm_string(s):
|
170 |
-
# Truncate if longer than 16 characters
|
171 |
-
if len(s) > 16:
|
172 |
-
return s[:16]
|
173 |
-
# Pad with zeros if shorter than 16 characters
|
174 |
-
else:
|
175 |
-
return s.ljust(16, '0')
|
176 |
-
def rhythm_string_to_array(s):
|
177 |
-
# Ensure the string is 16 characters long
|
178 |
-
s = s[:16].ljust(16, '0') # Truncate or pad with '0' to make it 16 characters
|
179 |
-
# Convert to numpy array, treating non-'0' as '1'
|
180 |
-
arr = np.array([1 if char != '0' else 0 for char in s], dtype=int)
|
181 |
-
arr = arr*np.array([3,1,2,1,3,1,2,1,3,1,2,1,3,1,2,1])
|
182 |
-
print(arr)
|
183 |
-
return arr
|
184 |
-
|
185 |
# Gradio main function
|
186 |
def generate_from_example(prompt):
|
187 |
-
midi_output, audio_output, midi = generate_music(prompt, tempo=80, mode="example", rhythm_control=
|
188 |
piano_roll_image = visualize_midi(midi)
|
189 |
return audio_output, piano_roll_image
|
190 |
|
191 |
-
def generate_diy(m1_chord, m2_chord, m3_chord, m4_chord,
|
192 |
-
m1_rhythm, m2_rhythm, m3_rhythm, m4_rhythm, tempo):
|
193 |
-
print("\n\n\n",m1_chord,type(m1_chord), "\n\n\n")
|
194 |
-
test_chd_roll = np.concatenate([np.tile(CHORD_DICTIONARY[m1_chord], (16, 1)),
|
195 |
-
np.tile(CHORD_DICTIONARY[m2_chord], (16, 1)),
|
196 |
-
np.tile(CHORD_DICTIONARY[m3_chord], (16, 1)),
|
197 |
-
np.tile(CHORD_DICTIONARY[m4_chord], (16, 1))])
|
198 |
-
rhythms = [m1_rhythm, m2_rhythm, m3_rhythm, m4_rhythm]
|
199 |
-
|
200 |
-
chd_roll = np.concatenate([test_chd_roll[np.newaxis,:,:], test_chd_roll[np.newaxis,:,:]], axis=0)
|
201 |
-
|
202 |
-
chd_roll = circular_extend(chd_roll)
|
203 |
-
chd_roll = -chd_roll-1
|
204 |
-
|
205 |
-
real_chd_roll = chd_roll
|
206 |
-
|
207 |
-
melody_roll = -np.ones_like(chd_roll)
|
208 |
-
|
209 |
-
if "null rhythm" not in rhythms:
|
210 |
-
rhythm_full = []
|
211 |
-
for i in range(len(rhythms)):
|
212 |
-
rhythm = adjust_rhythm_string(rhythms[i])
|
213 |
-
rhythm = rhythm_string_to_array(rhythm)
|
214 |
-
rhythm_full.append(rhythm)
|
215 |
-
rhythm_full = np.concatenate(rhythm_full, axis=0)
|
216 |
-
|
217 |
-
onset_roll = test_chd_roll*rhythm_full[:, np.newaxis]
|
218 |
-
sustain_roll = np.zeros_like(onset_roll)
|
219 |
-
no_onset_pos = np.all(onset_roll == 0, axis=-1)
|
220 |
-
sustain_roll[no_onset_pos] = test_chd_roll[no_onset_pos]
|
221 |
-
|
222 |
-
real_chd_roll = np.concatenate([onset_roll[np.newaxis,:,:], sustain_roll[np.newaxis,:,:]], axis=0)
|
223 |
-
real_chd_roll = circular_extend(real_chd_roll)
|
224 |
-
|
225 |
-
background_condition = np.concatenate([real_chd_roll, chd_roll, melody_roll], axis=0)
|
226 |
-
|
227 |
-
midi_output, audio_output, midi = generate_music(background_condition, tempo, mode="diy")
|
228 |
-
piano_roll_image = visualize_midi(midi)
|
229 |
-
return midi_output, audio_output, piano_roll_image
|
230 |
-
|
231 |
# Prompt list
|
232 |
prompt_list = ["example 1", "example 2", "example 3", "example 4"]
|
233 |
-
rhythm_list = ["null rhythm", "1010101010101010", "1011101010111010","1111101010111010","1010001010101010","1010101000101010"]
|
234 |
-
|
235 |
|
236 |
custom_css = """
|
237 |
-
.custom-row1 {
|
238 |
-
background-color: #fdebd0;
|
239 |
-
padding: 10px;
|
240 |
-
border-radius: 5px;
|
241 |
-
}
|
242 |
-
.custom-row2 {
|
243 |
-
background-color: #d1f2eb;
|
244 |
-
padding: 10px;
|
245 |
-
border-radius: 5px;
|
246 |
-
}
|
247 |
-
.custom-grey {
|
248 |
-
background-color: #f0f0f0;
|
249 |
-
padding: 10px;
|
250 |
-
border-radius: 5px;
|
251 |
-
}
|
252 |
.custom-purple {
|
253 |
background-color: #d7bde2;
|
254 |
padding: 10px;
|
@@ -259,227 +143,38 @@ custom_css = """
|
|
259 |
}
|
260 |
"""
|
261 |
|
262 |
-
|
263 |
with gr.Blocks(css=custom_css) as demo:
|
264 |
-
gr.Markdown("# <div style='text-align: center;font-size:40px'> Efficient Fine-Grained Guidance for Diffusion
|
265 |
|
266 |
-
gr.Markdown("<
|
267 |
-
|
268 |
-
\n   (2) Fine-grained control during the diffusion sampling process.\
|
269 |
-
\n In particular, **sampling control** ensures tonal accuracy in every generated sample, allowing our model to produce music with high precision, consistent rhythmic patterns,\
|
270 |
-
and even stylistic variations that align with user intent.<span>")
|
271 |
-
gr.Markdown("<span style='font-size:25px;color: red'> At the bottom of this page, we provide an interactive space for you to try our model by yourself! <span>")
|
272 |
|
|
|
|
|
|
|
|
|
273 |
|
274 |
-
gr.Markdown("\n\n\n")
|
275 |
-
gr.Markdown("# 1. Accompaniment Generation given Melody and Chord")
|
276 |
-
gr.Markdown("<span style='font-size:20px;'> In each example, the left column displays the melody provided as inputs to the model.\
|
277 |
-
The right column showcases music samples generated by the model.<span>")
|
278 |
-
|
279 |
-
with gr.Column(elem_classes="custom-row1"):
|
280 |
-
gr.Markdown("## Example 1")
|
281 |
-
with gr.Row():
|
282 |
-
with gr.Column():
|
283 |
-
gr.Markdown("<span style='font-size:20px;'> With the following melody as condition <span>")
|
284 |
-
example1_mel = gr.Audio(value="samples/diy_examples/example1/example_1_mel.wav", label="Melody", scale = 5)
|
285 |
-
with gr.Column():
|
286 |
-
gr.Markdown("<span style='font-size:20px;'> Generated Accompaniments <span>")
|
287 |
-
example1_audio = gr.Audio(value="samples/diy_examples/example1/sample1.wav", label="Generated Accompaniment", scale = 5)
|
288 |
-
|
289 |
-
with gr.Column(elem_classes="custom-row2"):
|
290 |
-
gr.Markdown("## Example 2")
|
291 |
-
with gr.Row():
|
292 |
-
with gr.Column():
|
293 |
-
gr.Markdown("<span style='font-size:20px;'> With the following melody as condition <span>")
|
294 |
-
example1_mel = gr.Audio(value="samples/diy_examples/example2/example_2_mel.wav", label="Melody", scale = 5)
|
295 |
-
with gr.Column():
|
296 |
-
gr.Markdown("<span style='font-size:20px;'> Generated Accompaniments <span>")
|
297 |
-
example1_audio = gr.Audio(value="samples/diy_examples/example2/sample1.wav", label="Generated Accompaniment", scale = 5)
|
298 |
-
|
299 |
-
with gr.Column(elem_classes="custom-row1"):
|
300 |
-
gr.Markdown("## Example 3")
|
301 |
-
with gr.Row():
|
302 |
-
with gr.Column():
|
303 |
-
gr.Markdown("<span style='font-size:20px;'> With the following melody as condition <span>")
|
304 |
-
example1_mel = gr.Audio(value="samples/diy_examples/example3/example_3_mel.wav", label="Melody", scale = 5)
|
305 |
-
with gr.Column():
|
306 |
-
gr.Markdown("<span style='font-size:20px;'> Generated Accompaniments <span>")
|
307 |
-
example1_audio = gr.Audio(value="samples/diy_examples/example3/sample1.wav", label="Generated Accompaniment", scale = 5)
|
308 |
-
|
309 |
-
with gr.Column(elem_classes="custom-row2"):
|
310 |
-
gr.Markdown("## Example 4")
|
311 |
-
with gr.Row():
|
312 |
-
with gr.Column():
|
313 |
-
gr.Markdown("<span style='font-size:20px;'> With the following melody as condition <span>")
|
314 |
-
example1_mel = gr.Audio(value="samples/diy_examples/example4/example_4_mel.wav", label="Melody", scale = 5)
|
315 |
-
with gr.Column():
|
316 |
-
gr.Markdown("<span style='font-size:20px;'> Generated Accompaniments <span>")
|
317 |
-
example1_audio = gr.Audio(value="samples/diy_examples/example4/sample1.wav", label="Generated Accompaniment", scale = 5)
|
318 |
-
|
319 |
-
gr.HTML("<div style='height: 50px;'></div>")
|
320 |
-
gr.Markdown("# \n\n\n")
|
321 |
-
gr.Markdown("# 2. Style-Controlled Music Generation")
|
322 |
-
gr.Markdown("<span style='font-size:20px;'>Our approach enables controllable stylization in music generation. The sampling control is able to\
|
323 |
-
ensure that all generated notes strictly adhere to the target musical style's scale.\
|
324 |
-
This allows the model to generate music in specific styles — even those that were not present in \
|
325 |
-
the training data.<span>")
|
326 |
-
gr.Markdown("<span style='font-size:20px;'> Below, we demonstrate several examples of style-controlled music generation for:\
|
327 |
-
\n   (1) Dorian Mode: (with scale being A-B-C-D-E-F#-G);\
|
328 |
-
\n   (2) Chinese Style: (with scale being C-D-E-G-A). <span>")
|
329 |
-
|
330 |
-
with gr.Column(elem_classes="custom-row1"):
|
331 |
-
gr.Markdown("## Dorian Mode")
|
332 |
-
gr.Markdown("<span style='font-size:20px;'> The following are two examples generated by our method <span>")
|
333 |
-
with gr.Row():
|
334 |
-
with gr.Column(elem_classes="custom-grey"):
|
335 |
-
gr.Markdown("<span style='font-size:20px;'> Example 1 <span>")
|
336 |
-
example1_mel = gr.Audio(value="samples/different_styles/dorian_1.wav", scale = 5)
|
337 |
-
with gr.Column(elem_classes="custom-grey"):
|
338 |
-
gr.Markdown("<span style='font-size:20px;'> Example 2 <span>")
|
339 |
-
example1_audio = gr.Audio(value="samples/different_styles/dorian_2.wav", scale = 5)
|
340 |
-
|
341 |
-
with gr.Column(elem_classes="custom-row2"):
|
342 |
-
gr.Markdown("## Chinese Style")
|
343 |
-
gr.Markdown("<span style='font-size:20px;'> The following are two examples generated by our method <span>")
|
344 |
-
with gr.Row():
|
345 |
-
with gr.Column(elem_classes="custom-grey"):
|
346 |
-
gr.Markdown("<span style='font-size:20px;'> Example 1 <span>")
|
347 |
-
example1_mel = gr.Audio(value="samples/different_styles/chinese_1.wav", scale = 5)
|
348 |
-
with gr.Column(elem_classes="custom-grey"):
|
349 |
-
gr.Markdown("<span style='font-size:20px;'> Example 2 <span>")
|
350 |
-
example1_audio = gr.Audio(value="samples/different_styles/chinese_2.wav", scale = 5)
|
351 |
-
|
352 |
gr.HTML("<div style='height: 50px;'></div>")
|
353 |
gr.Markdown("\n\n\n")
|
354 |
-
gr.Markdown("#
|
355 |
-
|
356 |
-
|
357 |
-
|
358 |
-
|
359 |
-
|
360 |
-
|
361 |
-
|
362 |
-
|
363 |
-
|
364 |
-
|
365 |
-
|
366 |
-
|
367 |
-
|
368 |
-
|
369 |
-
|
370 |
-
|
371 |
-
|
372 |
-
gr.Markdown("Melody Sheet")
|
373 |
-
example1_sheet = gr.Image(value="samples/control_vs_uncontrol/example_1_mel_chd.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
374 |
-
with gr.Column(scale=1, min_width=10, ):
|
375 |
-
gr.Markdown("Melody Audio")
|
376 |
-
example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_1_mel_chd.wav", label="Melody, wav", waveform_options=gr.WaveformOptions(show_recording_waveform=False), scale = 1, min_width=10)
|
377 |
-
|
378 |
-
gr.Markdown("## Generated Accompaniments")
|
379 |
-
with gr.Row(elem_classes="custom-grey"):
|
380 |
-
gr.Markdown("<span style='font-size:20px;'> Without sampling control<span>")
|
381 |
-
with gr.Column(scale=2, min_width=300):
|
382 |
-
gr.Markdown("Music Sheet")
|
383 |
-
example1_sheet = gr.Image(value="samples/control_vs_uncontrol/example_1_acc_uncontrol.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
384 |
-
with gr.Column(scale=1, min_width=150):
|
385 |
-
gr.Markdown("Audio")
|
386 |
-
example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_1_acc_uncontrol.wav", scale = 1, min_width=10)
|
387 |
-
gr.Markdown("\n\n\n")
|
388 |
-
with gr.Row(elem_classes="custom-grey"):
|
389 |
-
with gr.Column(scale=1, min_width=150):
|
390 |
-
gr.Markdown("<span style='font-size:20px;'>With sampling control<span>")
|
391 |
-
with gr.Column(scale=2, min_width=300):
|
392 |
-
gr.Markdown("Music Sheet")
|
393 |
-
example1_sheet = gr.Image(value="samples/control_vs_uncontrol/example_1_acc_control.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
394 |
-
with gr.Column(scale=1, min_width=150):
|
395 |
-
gr.Markdown("Audio")
|
396 |
-
example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_1_acc_control.wav", scale = 1, min_width=10)
|
397 |
-
|
398 |
-
|
399 |
-
with gr.Column(elem_classes="custom-row2"):
|
400 |
-
gr.Markdown("## Example 2")
|
401 |
-
|
402 |
-
with gr.Row(elem_classes="custom-grey"):
|
403 |
-
gr.Markdown("<span style='font-size:20px;'> With pre-defined melody and chord as follows<span>")
|
404 |
-
with gr.Column(scale=2, min_width=10, ):
|
405 |
-
gr.Markdown("Melody Sheet")
|
406 |
-
example1_sheet = gr.Image(value="samples/control_vs_uncontrol/example_2_mel_chd.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
407 |
-
with gr.Column(scale=1, min_width=10, ):
|
408 |
-
gr.Markdown("Melody Audio")
|
409 |
-
example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_2_mel_chd.wav", label="Melody, wav", waveform_options=gr.WaveformOptions(show_recording_waveform=False), scale = 1, min_width=10)
|
410 |
-
|
411 |
-
gr.Markdown("## Generated Accompaniments")
|
412 |
-
with gr.Row(elem_classes="custom-grey"):
|
413 |
-
gr.Markdown("<span style='font-size:20px;'> Without sampling control<span>")
|
414 |
-
with gr.Column(scale=2, min_width=300):
|
415 |
-
gr.Markdown("Music Sheet")
|
416 |
-
example1_sheet = gr.Image(value="samples/control_vs_uncontrol/example_2_acc_uncontrol.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
417 |
-
with gr.Column(scale=1, min_width=150):
|
418 |
-
gr.Markdown("Audio")
|
419 |
-
example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_2_acc_uncontrol.wav", scale = 1, min_width=10)
|
420 |
-
gr.Markdown("\n\n\n")
|
421 |
-
with gr.Row(elem_classes="custom-grey"):
|
422 |
-
with gr.Column(scale=1, min_width=150):
|
423 |
-
gr.Markdown("<span style='font-size:20px;'>With sampling control<span>")
|
424 |
-
with gr.Column(scale=2, min_width=300):
|
425 |
-
gr.Markdown("Music Sheet")
|
426 |
-
example1_sheet = gr.Image(value="samples/control_vs_uncontrol/example_2_acc_control.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
427 |
-
with gr.Column(scale=1, min_width=150):
|
428 |
-
gr.Markdown("Audio")
|
429 |
-
example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_2_acc_control.wav", scale = 1, min_width=10)
|
430 |
-
|
431 |
-
# with gr.Row():
|
432 |
-
# with gr.Column(scale=1, min_width=300, elem_classes="custom-row1"):
|
433 |
-
# gr.Markdown("## Example 1")
|
434 |
-
# gr.Markdown("<span style='font-size:20px;'> With pre-defined melody and chord as follows<span>")
|
435 |
-
# example1_sheet = gr.Image(value="samples/control_vs_uncontrol/example_1_mel_chd.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
436 |
-
# # Audio component to play the audio
|
437 |
-
# example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_1_mel_chd.wav", label="Melody, wav", waveform_options=gr.WaveformOptions(show_recording_waveform=False), scale = 1, min_width=10)
|
438 |
-
|
439 |
-
# gr.Markdown("## Generated Accompaniments")
|
440 |
-
# with gr.Row():
|
441 |
-
# with gr.Column(scale=1, min_width=150):
|
442 |
-
# gr.Markdown("<span style='font-size:20px;'> without sampling control<span>")
|
443 |
-
# example1_sheet = gr.Image(value="samples/control_vs_uncontrol/sample_1.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
444 |
-
# example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_1_acc_uncontrol.wav", scale = 1, min_width=10)
|
445 |
-
# with gr.Column(scale=1, min_width=150):
|
446 |
-
# gr.Markdown("<span style='font-size:20px;'> with sampling control<span>")
|
447 |
-
# example1_sheet = gr.Image(value="samples/control_vs_uncontrol/sample_1.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
448 |
-
# example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_1_acc_control.wav", scale = 1, min_width=10)
|
449 |
-
# with gr.Column(scale=1, min_width=300, elem_classes="custom-row2"):
|
450 |
-
# gr.Markdown("## Example 2")
|
451 |
-
# gr.Markdown("<span style='font-size:20px;'> With pre-defined melody and chord as follows<span>")
|
452 |
-
# example1_sheet = gr.Image(value="samples/control_vs_uncontrol/example_1_mel_chd.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
453 |
-
# # Audio component to play the audio
|
454 |
-
# example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_1_mel_chd.wav", label="Melody, wav", waveform_options=gr.WaveformOptions(show_recording_waveform=False), scale = 1, min_width=10)
|
455 |
-
|
456 |
-
# gr.Markdown("## Generated Accompaniments")
|
457 |
-
# with gr.Row():
|
458 |
-
# with gr.Column(scale=1, min_width=150):
|
459 |
-
# gr.Markdown("<span style='font-size:20px;'> without sampling control<span>")
|
460 |
-
# example1_sheet = gr.Image(value="samples/control_vs_uncontrol/sample_1.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
461 |
-
# example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_1_acc_uncontrol.wav", scale = 1, min_width=10)
|
462 |
-
# with gr.Column(scale=1, min_width=150):
|
463 |
-
# gr.Markdown("<span style='font-size:20px;'> with sampling control<span>")
|
464 |
-
# example1_sheet = gr.Image(value="samples/control_vs_uncontrol/sample_1.jpg", label="Music Sheet of Melody and Chord", scale=1, min_width=10)
|
465 |
-
# example1_melody = gr.Audio(value="samples/control_vs_uncontrol/example_1_acc_control.wav", scale = 1, min_width=10)
|
466 |
-
|
467 |
-
|
468 |
|
469 |
-
|
470 |
-
|
471 |
-
''' Try to generate by users '''
|
472 |
-
gr.HTML("<div style='height: 50px;'></div>")
|
473 |
-
gr.Markdown("\n\n\n")
|
474 |
-
gr.Markdown("# <span style='color: red;'> 4. DIY in real time! </span>")
|
475 |
-
gr.Markdown("<span style='font-size:20px;'> Here is an interactive tool for you to try our model and generate by yourself.\
|
476 |
-
You can generate new accompaniments for given melody and chord conditions <span>")
|
477 |
-
|
478 |
-
gr.Markdown("### <span style='color: blue;'> Currently this space is supported with Hugging Face CPU and on average,\
|
479 |
-
it takes about 15 seconds to generate a 4-measure music piece. However, if other users are generating\
|
480 |
-
music at the same time, one may enter a queue, which could slow down the process significantly.\
|
481 |
-
If that happens, feel free to refresh the page. We appreciate your patience and understanding.\
|
482 |
-
</span>")
|
483 |
|
484 |
with gr.Column(elem_classes="custom-purple"):
|
485 |
gr.Markdown("### Select an example to generate music given melody and chord condition")
|
@@ -502,7 +197,6 @@ with gr.Blocks(css=custom_css) as demo:
|
|
502 |
outputs=[audio_output, piano_roll_output]
|
503 |
)
|
504 |
|
505 |
-
|
506 |
# Launch Gradio interface
|
507 |
if __name__ == "__main__":
|
508 |
demo.launch()
|
|
|
57 |
elif selected_prompt == "example 6":
|
58 |
return "samples/diy_examples/example6/example6.jpg"
|
59 |
|
60 |
+
# Model for generating music
|
|
|
61 |
def generate_music(prompt, tempo, num_samples=1, mode="example", rhythm_control="Yes"):
|
|
|
62 |
ldm_model = init_ldm_model(params_chord_lsh_cond, debug_mode=False)
|
63 |
model = Diffpro_SDF.load_trained(ldm_model, model_path).to(device)
|
64 |
sampler = SDFSampler(model.ldm, 64, 64, is_autocast=False, device=device, debug_mode=False)
|
|
|
99 |
if background_condition[:,4:6,:,:].min()>=0:
|
100 |
full_lsh_roll = extend_piano_roll(background_condition[i,4:6,:,:].cpu().numpy())
|
101 |
midi_file = piano_roll_to_midi(full_roll, full_chd_roll, full_lsh_roll, bpm=tempo)
|
|
|
102 |
filename = f"output_{i}.mid"
|
103 |
save_midi(midi_file, filename)
|
104 |
subprocess.Popen(['timidity',f'output_{i}.mid','-Ow','-o',f'output_{i}.wav']).communicate()
|
|
|
123 |
plt.savefig(output_image_path)
|
124 |
return output_image_path
|
125 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
# Gradio main function
|
127 |
def generate_from_example(prompt):
|
128 |
+
midi_output, audio_output, midi = generate_music(prompt, tempo=80, mode="example", rhythm_control="No")
|
129 |
piano_roll_image = visualize_midi(midi)
|
130 |
return audio_output, piano_roll_image
|
131 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
132 |
# Prompt list
|
133 |
prompt_list = ["example 1", "example 2", "example 3", "example 4"]
|
|
|
|
|
134 |
|
135 |
custom_css = """
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
.custom-purple {
|
137 |
background-color: #d7bde2;
|
138 |
padding: 10px;
|
|
|
143 |
}
|
144 |
"""
|
145 |
|
|
|
146 |
with gr.Blocks(css=custom_css) as demo:
|
147 |
+
gr.Markdown("# <div style='text-align: center;font-size:40px'> Efficient Fine-Grained Guidance for Diffusion Model Based Symbolic Music Generation <div style='text-align: center;'>")
|
148 |
|
149 |
+
gr.Markdown("<div style='text-align: center;font-size:20px'>Tingyu Zhu<sup>*</sup>, Haoyu Liu<sup>*</sup>, Ziyu Wang, Zhimin Jiang, Zeyu Zheng</div>")
|
150 |
+
gr.Markdown("<div style='text-align: center;font-size:20px'><a href='https://arxiv.org/abs/2410.08435'>[Paper]</a> <a href='https://github.com/huajianduzhuo-code/FGG-music-code'>[Code Repo]</a></div>")
|
|
|
|
|
|
|
|
|
151 |
|
152 |
+
gr.Markdown("<span style='font-size:25px;'> For detailed information and demonstrations of our method, please visit our [GitHub Pages site](https://huajianduzhuo-code.github.io/FGG-diffusion-music/) to explore:\
|
153 |
+
\n   1. Accompaniment Generation given Melody and Chord\
|
154 |
+
\n   2. Style-Controlled Music Generation\
|
155 |
+
\n   3. Demonstrating the Effectiveness of Sampling Control by Comparison</span>")
|
156 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
157 |
gr.HTML("<div style='height: 50px;'></div>")
|
158 |
gr.Markdown("\n\n\n")
|
159 |
+
gr.Markdown("# <span style='color: red;'> Interactive Demo </span>")
|
160 |
+
gr.Markdown(
|
161 |
+
"<span style='font-size:20px;'>"
|
162 |
+
"🎵 Try out our interactive tool to generate music with our model!<br>"
|
163 |
+
"You can create new accompaniments conditioned on a given melody and chord progression."
|
164 |
+
"</span>"
|
165 |
+
)
|
166 |
+
|
167 |
+
gr.Markdown(
|
168 |
+
"<span style='color:blue; font-size:20px;'>"
|
169 |
+
"⚠️ This Space currently runs on a Hugging Face-provided CPU. On average, it takes ~15 seconds to generate a 4-measure music segment.<br>"
|
170 |
+
"If multiple users are generating at the same time, you may enter a queue, which can cause delays.<br><br>"
|
171 |
+
"🚀 On our local server (NVIDIA RTX 6000 Ada GPU), the same generation takes only 0.4 seconds.<br><br>"
|
172 |
+
"To speed things up, you can: <br>"
|
173 |
+
"• 🔁 Fork this Space and select a different hardware configuration<br>"
|
174 |
+
"• 🧑💻 Clone our <a href='https://github.com/huajianduzhuo-code/FGG-music-code'>[Code Repo]</a> and run the generation notebooks locally after installing dependencies and downloading the model weights."
|
175 |
+
"</span>"
|
176 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
177 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
178 |
|
179 |
with gr.Column(elem_classes="custom-purple"):
|
180 |
gr.Markdown("### Select an example to generate music given melody and chord condition")
|
|
|
197 |
outputs=[audio_output, piano_roll_output]
|
198 |
)
|
199 |
|
|
|
200 |
# Launch Gradio interface
|
201 |
if __name__ == "__main__":
|
202 |
demo.launch()
|
generation/__pycache__/gen_utils.cpython-39.pyc
CHANGED
Binary files a/generation/__pycache__/gen_utils.cpython-39.pyc and b/generation/__pycache__/gen_utils.cpython-39.pyc differ
|
|
model/__pycache__/__init__.cpython-39.pyc
CHANGED
Binary files a/model/__pycache__/__init__.cpython-39.pyc and b/model/__pycache__/__init__.cpython-39.pyc differ
|
|
model/__pycache__/latent_diffusion.cpython-39.pyc
CHANGED
Binary files a/model/__pycache__/latent_diffusion.cpython-39.pyc and b/model/__pycache__/latent_diffusion.cpython-39.pyc differ
|
|
model/__pycache__/model_sdf.cpython-39.pyc
CHANGED
Binary files a/model/__pycache__/model_sdf.cpython-39.pyc and b/model/__pycache__/model_sdf.cpython-39.pyc differ
|
|
model/__pycache__/sampler_sdf.cpython-39.pyc
CHANGED
Binary files a/model/__pycache__/sampler_sdf.cpython-39.pyc and b/model/__pycache__/sampler_sdf.cpython-39.pyc differ
|
|
model/architecture/__pycache__/unet.cpython-39.pyc
CHANGED
Binary files a/model/architecture/__pycache__/unet.cpython-39.pyc and b/model/architecture/__pycache__/unet.cpython-39.pyc differ
|
|
model/architecture/__pycache__/unet_attention.cpython-39.pyc
CHANGED
Binary files a/model/architecture/__pycache__/unet_attention.cpython-39.pyc and b/model/architecture/__pycache__/unet_attention.cpython-39.pyc differ
|
|
output_0.mid
CHANGED
Binary files a/output_0.mid and b/output_0.mid differ
|
|
output_0.wav
CHANGED
@@ -1,3 +1,3 @@
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
-
oid sha256:
|
3 |
-
size
|
|
|
1 |
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:15dcddbf23bcae2e04eb588ef0023e0e610d4b4f03709c5a777b4141a24d160e
|
3 |
+
size 2772208
|
piano_roll.png
CHANGED
![]() |
Git LFS Details
|
![]() |
Git LFS Details
|
train/__pycache__/__init__.cpython-39.pyc
CHANGED
Binary files a/train/__pycache__/__init__.cpython-39.pyc and b/train/__pycache__/__init__.cpython-39.pyc differ
|
|
train/__pycache__/learner.cpython-39.pyc
CHANGED
Binary files a/train/__pycache__/learner.cpython-39.pyc and b/train/__pycache__/learner.cpython-39.pyc differ
|
|
train/__pycache__/train_params.cpython-39.pyc
CHANGED
Binary files a/train/__pycache__/train_params.cpython-39.pyc and b/train/__pycache__/train_params.cpython-39.pyc differ
|
|