Update app.py
Browse files
app.py
CHANGED
@@ -24,8 +24,9 @@ import gradio as gr
|
|
24 |
from gradio import Markdown
|
25 |
from music21 import converter
|
26 |
import torchaudio.transforms as T
|
|
|
27 |
|
28 |
-
#
|
29 |
from utils import logger
|
30 |
from utils.btc_model import BTC_model
|
31 |
from utils.transformer_modules import *
|
@@ -38,20 +39,13 @@ from utils.mir_eval_modules import (
|
|
38 |
from utils.mert import FeatureExtractorMERT
|
39 |
from model.linear_mt_attn_ck import FeedforwardModelMTAttnCK
|
40 |
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
# Suppress unnecessary warnings and logs
|
45 |
warnings.filterwarnings("ignore")
|
46 |
logging.getLogger("transformers.modeling_utils").setLevel(logging.ERROR)
|
47 |
|
48 |
-
# from gradio import Markdown
|
49 |
-
|
50 |
PITCH_CLASS = ['C', 'C#', 'D', 'D#', 'E', 'F', 'F#', 'G', 'G#', 'A', 'A#', 'B']
|
51 |
-
|
52 |
tonic_signatures = ["A", "A#", "B", "C", "C#", "D", "D#", "E", "F", "F#", "G", "G#"]
|
53 |
-
mode_signatures = ["major", "minor"]
|
54 |
-
|
55 |
|
56 |
pitch_num_dic = {
|
57 |
'C': 0, 'C#': 1, 'D': 2, 'D#': 3, 'E': 4, 'F': 5,
|
@@ -97,20 +91,15 @@ def normalize_chord(file_path, key, key_type='major'):
|
|
97 |
new_key = "C major"
|
98 |
shift = 0
|
99 |
else:
|
100 |
-
#print ("asdas",key)
|
101 |
if len(key) == 1:
|
102 |
key = key[0].upper()
|
103 |
else:
|
104 |
key = key[0].upper() + key[1:]
|
105 |
-
|
106 |
if key in minor_major_dic2:
|
107 |
key = minor_major_dic2[key]
|
108 |
-
|
109 |
shift = 0
|
110 |
-
|
111 |
if key_type == "major":
|
112 |
new_key = "C major"
|
113 |
-
|
114 |
shift = shift_major_dic[key]
|
115 |
else:
|
116 |
new_key = "A minor"
|
@@ -118,31 +107,27 @@ def normalize_chord(file_path, key, key_type='major'):
|
|
118 |
|
119 |
converted_lines = []
|
120 |
for line in lines:
|
121 |
-
if line.strip():
|
122 |
parts = line.split()
|
123 |
start_time = parts[0]
|
124 |
end_time = parts[1]
|
125 |
-
chord = parts[2]
|
126 |
-
if chord == "N":
|
127 |
-
newchordnorm =
|
128 |
-
elif chord == "X":
|
129 |
-
newchordnorm = "X"
|
130 |
elif ":" in chord:
|
131 |
pitch = chord.split(":")[0]
|
132 |
attr = chord.split(":")[1]
|
133 |
-
pnum = pitch_num_dic
|
134 |
-
new_idx = (pnum - shift)%12
|
135 |
newchord = PITCH_CLASS[new_idx]
|
136 |
newchordnorm = newchord + ":" + attr
|
137 |
else:
|
138 |
pitch = chord
|
139 |
-
pnum = pitch_num_dic
|
140 |
-
new_idx = (pnum - shift)%12
|
141 |
newchord = PITCH_CLASS[new_idx]
|
142 |
newchordnorm = newchord
|
143 |
-
|
144 |
converted_lines.append(f"{start_time} {end_time} {newchordnorm}\n")
|
145 |
-
|
146 |
return converted_lines
|
147 |
|
148 |
def sanitize_key_signature(key):
|
@@ -157,146 +142,108 @@ def resample_waveform(waveform, original_sample_rate, target_sample_rate):
|
|
157 |
def split_audio(waveform, sample_rate):
|
158 |
segment_samples = segment_duration * sample_rate
|
159 |
total_samples = waveform.size(0)
|
160 |
-
|
161 |
segments = []
|
162 |
for start in range(0, total_samples, segment_samples):
|
163 |
end = start + segment_samples
|
164 |
if end <= total_samples:
|
165 |
-
|
166 |
-
segments.append(segment)
|
167 |
-
|
168 |
-
# In case audio length is shorter than segment length.
|
169 |
if len(segments) == 0:
|
170 |
-
|
171 |
-
segments.append(segment)
|
172 |
-
|
173 |
return segments
|
174 |
|
175 |
-
|
176 |
def safe_remove_dir(directory):
|
177 |
-
"""
|
178 |
-
Safely removes a directory only if it exists and is empty.
|
179 |
-
"""
|
180 |
directory = Path(directory)
|
181 |
if directory.exists():
|
182 |
try:
|
183 |
shutil.rmtree(directory)
|
184 |
-
except FileNotFoundError:
|
185 |
-
print(f"Warning: Some files in {directory} were already deleted.")
|
186 |
-
except PermissionError:
|
187 |
-
print(f"Warning: Permission issue encountered while deleting {directory}.")
|
188 |
except Exception as e:
|
189 |
-
print(f"
|
190 |
-
|
191 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
192 |
class Music2emo:
|
193 |
-
def __init__(
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
|
198 |
-
local_files_only=False,
|
199 |
-
):
|
200 |
-
|
201 |
-
# use_cuda = torch.cuda.is_available()
|
202 |
-
# self.device = torch.device("cuda" if use_cuda else "cpu")
|
203 |
model_weights = "saved_models/J_all.ckpt"
|
204 |
self.device = device
|
205 |
-
|
206 |
self.feature_extractor = FeatureExtractorMERT(model_name='m-a-p/MERT-v1-95M', device=self.device, sr=resample_rate)
|
207 |
self.model_weights = model_weights
|
208 |
-
|
209 |
self.music2emo_model = FeedforwardModelMTAttnCK(
|
210 |
-
input_size=
|
211 |
output_size_classification=56,
|
212 |
output_size_regression=2
|
213 |
)
|
214 |
-
|
215 |
checkpoint = torch.load(self.model_weights, map_location=self.device, weights_only=False)
|
216 |
-
state_dict = checkpoint["state_dict"]
|
217 |
-
|
218 |
-
# Adjust the keys in the state_dict
|
219 |
-
state_dict = {key.replace("model.", ""): value for key, value in state_dict.items()}
|
220 |
-
|
221 |
-
# Filter state_dict to match model's keys
|
222 |
model_keys = set(self.music2emo_model.state_dict().keys())
|
223 |
filtered_state_dict = {key: value for key, value in state_dict.items() if key in model_keys}
|
224 |
-
|
225 |
-
# Load the filtered state_dict and set the model to evaluation mode
|
226 |
self.music2emo_model.load_state_dict(filtered_state_dict)
|
227 |
-
|
228 |
self.music2emo_model.to(self.device)
|
229 |
self.music2emo_model.eval()
|
230 |
-
|
231 |
self.config = HParams.load("./inference/data/run_config.yaml")
|
232 |
self.config.feature['large_voca'] = True
|
233 |
self.config.model['num_chords'] = 170
|
234 |
model_file = './inference/data/btc_model_large_voca.pt'
|
235 |
self.idx_to_voca = idx2voca_chord()
|
236 |
self.btc_model = BTC_model(config=self.config.model).to(self.device)
|
237 |
-
|
238 |
if os.path.isfile(model_file):
|
239 |
checkpoint = torch.load(model_file, map_location=self.device)
|
240 |
self.mean = checkpoint['mean']
|
241 |
self.std = checkpoint['std']
|
242 |
self.btc_model.load_state_dict(checkpoint['model'])
|
243 |
-
|
244 |
-
|
245 |
self.tonic_to_idx = {tonic: idx for idx, tonic in enumerate(tonic_signatures)}
|
246 |
self.mode_to_idx = {mode: idx for idx, mode in enumerate(mode_signatures)}
|
247 |
self.idx_to_tonic = {idx: tonic for tonic, idx in self.tonic_to_idx.items()}
|
248 |
self.idx_to_mode = {idx: mode for mode, idx in self.mode_to_idx.items()}
|
249 |
-
|
250 |
with open('inference/data/chord.json', 'r') as f:
|
251 |
self.chord_to_idx = json.load(f)
|
252 |
with open('inference/data/chord_inv.json', 'r') as f:
|
253 |
-
self.idx_to_chord = json.load(f)
|
254 |
-
self.idx_to_chord = {int(k): v for k, v in self.idx_to_chord.items()} # Ensure keys are ints
|
255 |
with open('inference/data/chord_root.json') as json_file:
|
256 |
self.chordRootDic = json.load(json_file)
|
257 |
with open('inference/data/chord_attr.json') as json_file:
|
258 |
self.chordAttrDic = json.load(json_file)
|
259 |
|
260 |
-
|
261 |
-
|
262 |
-
def predict(self, audio, threshold = 0.5):
|
263 |
-
|
264 |
feature_dir = Path("./inference/temp_out")
|
265 |
output_dir = Path("./inference/output")
|
266 |
-
|
267 |
-
# if feature_dir.exists():
|
268 |
-
# shutil.rmtree(str(feature_dir))
|
269 |
-
# if output_dir.exists():
|
270 |
-
# shutil.rmtree(str(output_dir))
|
271 |
-
|
272 |
-
# feature_dir.mkdir(parents=True)
|
273 |
-
# output_dir.mkdir(parents=True)
|
274 |
-
|
275 |
-
# warnings.filterwarnings('ignore')
|
276 |
-
# logger.logging_verbosity(1)
|
277 |
-
|
278 |
-
# mert_dir = feature_dir / "mert"
|
279 |
-
# mert_dir.mkdir(parents=True)
|
280 |
-
|
281 |
safe_remove_dir(feature_dir)
|
282 |
safe_remove_dir(output_dir)
|
283 |
-
|
284 |
feature_dir.mkdir(parents=True, exist_ok=True)
|
285 |
output_dir.mkdir(parents=True, exist_ok=True)
|
286 |
-
|
287 |
warnings.filterwarnings('ignore')
|
288 |
logger.logging_verbosity(1)
|
289 |
-
|
290 |
mert_dir = feature_dir / "mert"
|
291 |
mert_dir.mkdir(parents=True, exist_ok=True)
|
292 |
-
|
293 |
waveform, sample_rate = torchaudio.load(audio)
|
294 |
if waveform.shape[0] > 1:
|
295 |
waveform = waveform.mean(dim=0).unsqueeze(0)
|
296 |
waveform = waveform.squeeze()
|
297 |
waveform, sample_rate = resample_waveform(waveform, sample_rate, resample_rate)
|
298 |
-
|
299 |
-
if is_split:
|
300 |
segments = split_audio(waveform, sample_rate)
|
301 |
for i, segment in enumerate(segments):
|
302 |
segment_save_path = os.path.join(mert_dir, f"segment_{i}.npy")
|
@@ -304,50 +251,38 @@ class Music2emo:
|
|
304 |
else:
|
305 |
segment_save_path = os.path.join(mert_dir, f"segment_0.npy")
|
306 |
self.feature_extractor.extract_features_from_segment(waveform, sample_rate, segment_save_path)
|
307 |
-
|
308 |
-
embeddings = []
|
309 |
-
layers_to_extract = [5,6]
|
310 |
segment_embeddings = []
|
311 |
-
|
|
|
312 |
file_path = os.path.join(mert_dir, filename)
|
313 |
if os.path.isfile(file_path) and filename.endswith('.npy'):
|
314 |
segment = np.load(file_path)
|
315 |
concatenated_features = np.concatenate(
|
316 |
[segment[:, layer_idx, :] for layer_idx in layers_to_extract], axis=1
|
317 |
)
|
318 |
-
concatenated_features = np.squeeze(concatenated_features)
|
319 |
segment_embeddings.append(concatenated_features)
|
320 |
-
|
321 |
segment_embeddings = np.array(segment_embeddings)
|
322 |
if len(segment_embeddings) > 0:
|
323 |
final_embedding_mert = np.mean(segment_embeddings, axis=0)
|
324 |
else:
|
325 |
final_embedding_mert = np.zeros((1536,))
|
326 |
-
|
327 |
-
final_embedding_mert = torch.from_numpy(final_embedding_mert)
|
328 |
-
final_embedding_mert.to(self.device)
|
329 |
-
|
330 |
-
# --- Chord feature extract ---
|
331 |
-
|
332 |
audio_path = audio
|
333 |
-
audio_id =
|
334 |
try:
|
335 |
feature, feature_per_second, song_length_second = audio_file_to_features(audio_path, self.config)
|
336 |
except:
|
337 |
-
logger.info("
|
338 |
assert(False)
|
339 |
-
|
340 |
-
logger.info("audio file loaded and feature computation success : %s" % audio_path)
|
341 |
-
|
342 |
feature = feature.T
|
343 |
feature = (feature - self.mean) / self.std
|
344 |
time_unit = feature_per_second
|
345 |
n_timestep = self.config.model['timestep']
|
346 |
-
|
347 |
num_pad = n_timestep - (feature.shape[0] % n_timestep)
|
348 |
feature = np.pad(feature, ((0, num_pad), (0, 0)), mode="constant", constant_values=0)
|
349 |
num_instance = feature.shape[0] // n_timestep
|
350 |
-
|
351 |
start_time = 0.0
|
352 |
lines = []
|
353 |
with torch.no_grad():
|
@@ -362,85 +297,30 @@ class Music2emo:
|
|
362 |
prev_chord = prediction[i].item()
|
363 |
continue
|
364 |
if prediction[i].item() != prev_chord:
|
365 |
-
lines.append(
|
366 |
-
'%.3f %.3f %s\n' % (start_time, time_unit * (n_timestep * t + i), self.idx_to_voca[prev_chord]))
|
367 |
start_time = time_unit * (n_timestep * t + i)
|
368 |
prev_chord = prediction[i].item()
|
369 |
if t == num_instance - 1 and i + num_pad == n_timestep:
|
370 |
if start_time != time_unit * (n_timestep * t + i):
|
371 |
lines.append('%.3f %.3f %s\n' % (start_time, time_unit * (n_timestep * t + i), self.idx_to_voca[prev_chord]))
|
372 |
break
|
373 |
-
|
374 |
save_path = os.path.join(feature_dir, os.path.split(audio_path)[-1].replace('.mp3', '').replace('.wav', '') + '.lab')
|
375 |
with open(save_path, 'w') as f:
|
376 |
for line in lines:
|
377 |
f.write(line)
|
378 |
-
|
379 |
-
# logger.info("label file saved : %s" % save_path)
|
380 |
-
|
381 |
-
# lab file to midi file
|
382 |
-
starts, ends, pitchs = list(), list(), list()
|
383 |
-
|
384 |
-
intervals, chords = mir_eval.io.load_labeled_intervals(save_path)
|
385 |
-
for p in range(12):
|
386 |
-
for i, (interval, chord) in enumerate(zip(intervals, chords)):
|
387 |
-
root_num, relative_bitmap, _ = mir_eval.chord.encode(chord)
|
388 |
-
tmp_label = mir_eval.chord.rotate_bitmap_to_root(relative_bitmap, root_num)[p]
|
389 |
-
if i == 0:
|
390 |
-
start_time = interval[0]
|
391 |
-
label = tmp_label
|
392 |
-
continue
|
393 |
-
if tmp_label != label:
|
394 |
-
if label == 1.0:
|
395 |
-
starts.append(start_time), ends.append(interval[0]), pitchs.append(p + 48)
|
396 |
-
start_time = interval[0]
|
397 |
-
label = tmp_label
|
398 |
-
if i == (len(intervals) - 1):
|
399 |
-
if label == 1.0:
|
400 |
-
starts.append(start_time), ends.append(interval[1]), pitchs.append(p + 48)
|
401 |
-
|
402 |
-
midi = pm.PrettyMIDI()
|
403 |
-
instrument = pm.Instrument(program=0)
|
404 |
-
|
405 |
-
for start, end, pitch in zip(starts, ends, pitchs):
|
406 |
-
pm_note = pm.Note(velocity=120, pitch=pitch, start=start, end=end)
|
407 |
-
instrument.notes.append(pm_note)
|
408 |
-
|
409 |
-
midi.instruments.append(instrument)
|
410 |
-
midi.write(save_path.replace('.lab', '.midi'))
|
411 |
-
|
412 |
-
|
413 |
-
|
414 |
-
|
415 |
try:
|
416 |
midi_file = converter.parse(save_path.replace('.lab', '.midi'))
|
417 |
key_signature = str(midi_file.analyze('key'))
|
418 |
except Exception as e:
|
419 |
key_signature = "None"
|
420 |
-
|
421 |
key_parts = key_signature.split()
|
422 |
-
key_signature = sanitize_key_signature(key_parts[0])
|
423 |
key_type = key_parts[1] if len(key_parts) > 1 else 'major'
|
424 |
-
|
425 |
-
# --- Key feature (Tonic and Mode separation) ---
|
426 |
-
if key_signature == "None":
|
427 |
-
mode = "major"
|
428 |
-
else:
|
429 |
-
mode = key_signature.split()[-1]
|
430 |
-
|
431 |
-
encoded_mode = self.mode_to_idx.get(mode, 0)
|
432 |
-
mode_tensor = torch.tensor([encoded_mode], dtype=torch.long).to(self.device)
|
433 |
-
|
434 |
converted_lines = normalize_chord(save_path, key_signature, key_type)
|
435 |
-
|
436 |
lab_norm_path = save_path[:-4] + "_norm.lab"
|
437 |
-
|
438 |
-
# Write the converted lines to the new file
|
439 |
with open(lab_norm_path, 'w') as f:
|
440 |
f.writelines(converted_lines)
|
441 |
-
|
442 |
chords = []
|
443 |
-
|
444 |
if not os.path.exists(lab_norm_path):
|
445 |
chords.append((float(0), float(0), "N"))
|
446 |
else:
|
@@ -448,202 +328,148 @@ class Music2emo:
|
|
448 |
for line in file:
|
449 |
start, end, chord = line.strip().split()
|
450 |
chords.append((float(start), float(end), chord))
|
451 |
-
|
452 |
encoded = []
|
453 |
-
encoded_root= []
|
454 |
-
encoded_attr=[]
|
455 |
durations = []
|
456 |
-
|
457 |
for start, end, chord in chords:
|
458 |
chord_arr = chord.split(":")
|
459 |
if len(chord_arr) == 1:
|
460 |
chordRootID = self.chordRootDic[chord_arr[0]]
|
461 |
-
if chord_arr[0]
|
462 |
-
chordAttrID = 0
|
463 |
-
else:
|
464 |
-
chordAttrID = 1
|
465 |
elif len(chord_arr) == 2:
|
466 |
chordRootID = self.chordRootDic[chord_arr[0]]
|
467 |
chordAttrID = self.chordAttrDic[chord_arr[1]]
|
468 |
encoded_root.append(chordRootID)
|
469 |
encoded_attr.append(chordAttrID)
|
470 |
-
|
471 |
if chord in self.chord_to_idx:
|
472 |
encoded.append(self.chord_to_idx[chord])
|
473 |
else:
|
474 |
-
print(f"
|
475 |
-
|
476 |
-
durations.append(end - start) # Compute duration
|
477 |
-
|
478 |
encoded_chords = np.array(encoded)
|
479 |
encoded_chords_root = np.array(encoded_root)
|
480 |
encoded_chords_attr = np.array(encoded_attr)
|
481 |
-
|
482 |
-
# Maximum sequence length for chords
|
483 |
-
max_sequence_length = 100 # Define this globally or as a parameter
|
484 |
-
|
485 |
-
# Truncate or pad chord sequences
|
486 |
if len(encoded_chords) > max_sequence_length:
|
487 |
-
# Truncate to max length
|
488 |
encoded_chords = encoded_chords[:max_sequence_length]
|
489 |
encoded_chords_root = encoded_chords_root[:max_sequence_length]
|
490 |
encoded_chords_attr = encoded_chords_attr[:max_sequence_length]
|
491 |
-
|
492 |
else:
|
493 |
-
# Pad with zeros (padding value for chords)
|
494 |
padding = [0] * (max_sequence_length - len(encoded_chords))
|
495 |
encoded_chords = np.concatenate([encoded_chords, padding])
|
496 |
encoded_chords_root = np.concatenate([encoded_chords_root, padding])
|
497 |
encoded_chords_attr = np.concatenate([encoded_chords_attr, padding])
|
498 |
-
|
499 |
-
# Convert to tensor
|
500 |
chords_tensor = torch.tensor(encoded_chords, dtype=torch.long).to(self.device)
|
501 |
chords_root_tensor = torch.tensor(encoded_chords_root, dtype=torch.long).to(self.device)
|
502 |
chords_attr_tensor = torch.tensor(encoded_chords_attr, dtype=torch.long).to(self.device)
|
503 |
-
|
504 |
model_input_dic = {
|
505 |
"x_mert": final_embedding_mert.unsqueeze(0),
|
506 |
"x_chord": chords_tensor.unsqueeze(0),
|
507 |
"x_chord_root": chords_root_tensor.unsqueeze(0),
|
508 |
"x_chord_attr": chords_attr_tensor.unsqueeze(0),
|
509 |
-
"x_key":
|
510 |
}
|
511 |
-
|
512 |
model_input_dic = {k: v.to(self.device) for k, v in model_input_dic.items()}
|
513 |
classification_output, regression_output = self.music2emo_model(model_input_dic)
|
514 |
-
|
515 |
-
|
516 |
-
tag_list = np.load ( "./inference/data/tag_list.npy")
|
517 |
tag_list = tag_list[127:]
|
518 |
mood_list = [t.replace("mood/theme---", "") for t in tag_list]
|
519 |
-
threshold = threshold
|
520 |
-
|
521 |
-
# Get probabilities
|
522 |
probs = torch.sigmoid(classification_output).squeeze().tolist()
|
523 |
-
|
524 |
-
# Include both mood names and scores
|
525 |
predicted_moods_with_scores = [
|
526 |
-
{"mood": mood_list[i], "score": round(p, 4)}
|
527 |
for i, p in enumerate(probs) if p > threshold
|
528 |
]
|
529 |
-
|
530 |
-
# Include both mood names and scores
|
531 |
predicted_moods_with_scores_all = [
|
532 |
-
{"mood": mood_list[i], "score": round(p, 4)}
|
533 |
for i, p in enumerate(probs)
|
534 |
]
|
535 |
-
|
536 |
-
|
537 |
-
# Sort by highest probability
|
538 |
predicted_moods_with_scores.sort(key=lambda x: x["score"], reverse=True)
|
539 |
-
|
540 |
valence, arousal = regression_output.squeeze().tolist()
|
541 |
-
|
542 |
model_output_dic = {
|
543 |
"valence": valence,
|
544 |
"arousal": arousal,
|
545 |
"predicted_moods": predicted_moods_with_scores,
|
546 |
"predicted_moods_all": predicted_moods_with_scores_all
|
547 |
}
|
548 |
-
|
549 |
-
# predicted_moods = [mood_list[i] for i, p in enumerate(probs.squeeze().tolist()) if p > threshold]
|
550 |
-
# valence, arousal = regression_output.squeeze().tolist()
|
551 |
-
# model_output_dic = {
|
552 |
-
# "valence": valence,
|
553 |
-
# "arousal": arousal,
|
554 |
-
# "predicted_moods": predicted_moods
|
555 |
-
# }
|
556 |
-
|
557 |
return model_output_dic
|
558 |
|
559 |
-
# Music2Emo
|
560 |
if torch.cuda.is_available():
|
561 |
music2emo = Music2emo()
|
562 |
else:
|
563 |
music2emo = Music2emo(device="cpu")
|
564 |
|
565 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
566 |
def plot_mood_probabilities(predicted_moods_with_scores):
|
567 |
-
"""Plot mood probabilities as a horizontal bar chart."""
|
568 |
if not predicted_moods_with_scores:
|
569 |
return None
|
570 |
-
|
571 |
-
# Extract mood names and their scores
|
572 |
moods = [m["mood"] for m in predicted_moods_with_scores]
|
573 |
probs = [m["score"] for m in predicted_moods_with_scores]
|
574 |
-
|
575 |
-
# Sort moods by probability
|
576 |
sorted_indices = np.argsort(probs)[::-1]
|
577 |
sorted_probs = [probs[i] for i in sorted_indices]
|
578 |
sorted_moods = [moods[i] for i in sorted_indices]
|
579 |
-
|
580 |
-
# Create bar chart
|
581 |
fig, ax = plt.subplots(figsize=(8, 4))
|
582 |
ax.barh(sorted_moods[:10], sorted_probs[:10], color="#4CAF50")
|
583 |
-
ax.set_xlabel("
|
584 |
-
ax.set_title("
|
585 |
ax.invert_yaxis()
|
586 |
-
|
587 |
return fig
|
588 |
|
589 |
def plot_valence_arousal(valence, arousal):
|
590 |
-
"""Plot valence-arousal on a 2D circumplex model."""
|
591 |
fig, ax = plt.subplots(figsize=(4, 4))
|
592 |
ax.scatter(valence, arousal, color="red", s=100)
|
593 |
ax.set_xlim(1, 9)
|
594 |
ax.set_ylim(1, 9)
|
595 |
-
|
596 |
-
|
597 |
-
ax.
|
598 |
-
ax.
|
599 |
-
|
600 |
-
# Labels & Grid
|
601 |
-
ax.set_xlabel("Valence (Positivity)")
|
602 |
-
ax.set_ylabel("Arousal (Intensity)")
|
603 |
-
ax.set_title("Valence-Arousal Plot")
|
604 |
-
ax.legend()
|
605 |
ax.grid(True, linestyle="--", alpha=0.6)
|
606 |
-
|
607 |
return fig
|
608 |
|
609 |
-
|
610 |
-
|
611 |
-
def format_prediction(model_output_dic):
|
612 |
-
"""Format the model output in a structured format"""
|
613 |
-
valence = model_output_dic["valence"]
|
614 |
-
arousal = model_output_dic["arousal"]
|
615 |
-
predicted_moods_with_scores = model_output_dic["predicted_moods"]
|
616 |
-
predicted_moods_with_scores_all = model_output_dic["predicted_moods_all"]
|
617 |
-
|
618 |
-
# Generate charts
|
619 |
-
va_chart = plot_valence_arousal(valence, arousal)
|
620 |
-
mood_chart = plot_mood_probabilities(predicted_moods_with_scores_all)
|
621 |
-
|
622 |
-
# Format mood output with scores
|
623 |
-
if predicted_moods_with_scores:
|
624 |
-
moods_text = ", ".join(
|
625 |
-
[f"{m['mood']} ({m['score']:.2f})" for m in predicted_moods_with_scores]
|
626 |
-
)
|
627 |
-
else:
|
628 |
-
moods_text = "No significant moods detected."
|
629 |
-
|
630 |
-
# Create formatted output
|
631 |
-
output_text = f"""🎭 Predicted Mood Tags: {moods_text}
|
632 |
-
|
633 |
-
💖 Valence: {valence:.2f} (Scale: 1-9)
|
634 |
-
⚡ Arousal: {arousal:.2f} (Scale: 1-9)"""
|
635 |
-
|
636 |
-
return output_text, va_chart, mood_chart
|
637 |
-
|
638 |
-
# Gradio UI Elements
|
639 |
-
title="🎵 Music2Emo: Toward Unified Music Emotion Recognition"
|
640 |
description_text = """
|
641 |
-
<p>
|
642 |
-
|
|
|
|
|
643 |
</p>
|
644 |
"""
|
645 |
-
|
646 |
-
# Custom CSS Styling
|
647 |
css = """
|
648 |
.gradio-container {
|
649 |
font-family: 'Inter', -apple-system, system-ui, sans-serif;
|
@@ -654,7 +480,6 @@ css = """
|
|
654 |
border-radius: 8px;
|
655 |
padding: 10px;
|
656 |
}
|
657 |
-
/* Add padding to the top of the two plot boxes */
|
658 |
.gr-box {
|
659 |
padding-top: 25px !important;
|
660 |
}
|
@@ -663,52 +488,27 @@ css = """
|
|
663 |
with gr.Blocks(css=css) as demo:
|
664 |
gr.HTML(f"<h1 style='text-align: center;'>{title}</h1>")
|
665 |
gr.Markdown(description_text)
|
666 |
-
|
667 |
-
# Notes Section
|
668 |
gr.Markdown("""
|
669 |
-
### 📝
|
670 |
-
-
|
671 |
-
- **
|
|
|
672 |
""")
|
673 |
-
|
674 |
with gr.Row():
|
675 |
-
# Left Panel (Input)
|
676 |
with gr.Column(scale=1):
|
677 |
-
input_audio = gr.Audio(
|
678 |
-
|
679 |
-
|
680 |
-
)
|
681 |
-
threshold = gr.Slider(
|
682 |
-
minimum=0.0,
|
683 |
-
maximum=1.0,
|
684 |
-
value=0.5,
|
685 |
-
step=0.01,
|
686 |
-
label="Mood Detection Threshold",
|
687 |
-
info="Adjust threshold for mood detection"
|
688 |
-
)
|
689 |
-
predict_btn = gr.Button("🎭 Analyze Emotions", variant="primary")
|
690 |
-
|
691 |
-
# Right Panel (Output)
|
692 |
with gr.Column(scale=1):
|
693 |
-
output_text = gr.Textbox(
|
694 |
-
label="Analysis Results",
|
695 |
-
lines=4,
|
696 |
-
interactive=False # Prevent user input
|
697 |
-
)
|
698 |
-
|
699 |
-
# Ensure both plots have padding on top
|
700 |
with gr.Row(equal_height=True):
|
701 |
-
mood_chart = gr.Plot(label="
|
702 |
-
va_chart = gr.Plot(label="
|
703 |
-
|
704 |
predict_btn.click(
|
705 |
-
fn=
|
706 |
-
inputs=[input_audio, threshold],
|
707 |
outputs=[output_text, va_chart, mood_chart]
|
708 |
)
|
709 |
|
710 |
-
# Launch the App
|
711 |
demo.queue().launch()
|
712 |
-
|
713 |
-
|
714 |
-
|
|
|
24 |
from gradio import Markdown
|
25 |
from music21 import converter
|
26 |
import torchaudio.transforms as T
|
27 |
+
import matplotlib.pyplot as plt
|
28 |
|
29 |
+
# カスタムユーティリティのインポート
|
30 |
from utils import logger
|
31 |
from utils.btc_model import BTC_model
|
32 |
from utils.transformer_modules import *
|
|
|
39 |
from utils.mert import FeatureExtractorMERT
|
40 |
from model.linear_mt_attn_ck import FeedforwardModelMTAttnCK
|
41 |
|
42 |
+
# 不要な警告・ログを抑制
|
|
|
|
|
|
|
43 |
warnings.filterwarnings("ignore")
|
44 |
logging.getLogger("transformers.modeling_utils").setLevel(logging.ERROR)
|
45 |
|
|
|
|
|
46 |
PITCH_CLASS = ['C', 'C#', 'D', 'D#', 'E', 'F', 'F#', 'G', 'G#', 'A', 'A#', 'B']
|
|
|
47 |
tonic_signatures = ["A", "A#", "B", "C", "C#", "D", "D#", "E", "F", "F#", "G", "G#"]
|
48 |
+
mode_signatures = ["major", "minor"]
|
|
|
49 |
|
50 |
pitch_num_dic = {
|
51 |
'C': 0, 'C#': 1, 'D': 2, 'D#': 3, 'E': 4, 'F': 5,
|
|
|
91 |
new_key = "C major"
|
92 |
shift = 0
|
93 |
else:
|
|
|
94 |
if len(key) == 1:
|
95 |
key = key[0].upper()
|
96 |
else:
|
97 |
key = key[0].upper() + key[1:]
|
|
|
98 |
if key in minor_major_dic2:
|
99 |
key = minor_major_dic2[key]
|
|
|
100 |
shift = 0
|
|
|
101 |
if key_type == "major":
|
102 |
new_key = "C major"
|
|
|
103 |
shift = shift_major_dic[key]
|
104 |
else:
|
105 |
new_key = "A minor"
|
|
|
107 |
|
108 |
converted_lines = []
|
109 |
for line in lines:
|
110 |
+
if line.strip():
|
111 |
parts = line.split()
|
112 |
start_time = parts[0]
|
113 |
end_time = parts[1]
|
114 |
+
chord = parts[2]
|
115 |
+
if chord == "N" or chord == "X":
|
116 |
+
newchordnorm = chord
|
|
|
|
|
117 |
elif ":" in chord:
|
118 |
pitch = chord.split(":")[0]
|
119 |
attr = chord.split(":")[1]
|
120 |
+
pnum = pitch_num_dic[pitch]
|
121 |
+
new_idx = (pnum - shift) % 12
|
122 |
newchord = PITCH_CLASS[new_idx]
|
123 |
newchordnorm = newchord + ":" + attr
|
124 |
else:
|
125 |
pitch = chord
|
126 |
+
pnum = pitch_num_dic[pitch]
|
127 |
+
new_idx = (pnum - shift) % 12
|
128 |
newchord = PITCH_CLASS[new_idx]
|
129 |
newchordnorm = newchord
|
|
|
130 |
converted_lines.append(f"{start_time} {end_time} {newchordnorm}\n")
|
|
|
131 |
return converted_lines
|
132 |
|
133 |
def sanitize_key_signature(key):
|
|
|
142 |
def split_audio(waveform, sample_rate):
|
143 |
segment_samples = segment_duration * sample_rate
|
144 |
total_samples = waveform.size(0)
|
|
|
145 |
segments = []
|
146 |
for start in range(0, total_samples, segment_samples):
|
147 |
end = start + segment_samples
|
148 |
if end <= total_samples:
|
149 |
+
segments.append(waveform[start:end])
|
|
|
|
|
|
|
150 |
if len(segments) == 0:
|
151 |
+
segments.append(waveform)
|
|
|
|
|
152 |
return segments
|
153 |
|
|
|
154 |
def safe_remove_dir(directory):
|
|
|
|
|
|
|
155 |
directory = Path(directory)
|
156 |
if directory.exists():
|
157 |
try:
|
158 |
shutil.rmtree(directory)
|
|
|
|
|
|
|
|
|
159 |
except Exception as e:
|
160 |
+
print(f"ディレクトリ {directory} の削除中にエラーが発生しました: {e}")
|
161 |
+
|
162 |
+
# 追加:YouTube URL から音声をダウンロードする関数
|
163 |
+
def download_audio_from_youtube(url, output_dir="inference/input"):
|
164 |
+
import yt_dlp
|
165 |
+
os.makedirs(output_dir, exist_ok=True)
|
166 |
+
ydl_opts = {
|
167 |
+
'format': 'bestaudio/best',
|
168 |
+
'outtmpl': os.path.join(output_dir, 'tmp.%(ext)s'),
|
169 |
+
'postprocessors': [{
|
170 |
+
'key': 'FFmpegExtractAudio',
|
171 |
+
'preferredcodec': 'mp3',
|
172 |
+
'preferredquality': '192',
|
173 |
+
}],
|
174 |
+
'noplaylist': True,
|
175 |
+
'quiet': True,
|
176 |
+
}
|
177 |
+
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
|
178 |
+
info = ydl.extract_info(url, download=True)
|
179 |
+
title = info.get('title', '不明なタイトル')
|
180 |
+
output_file = os.path.join(output_dir, 'tmp.mp3')
|
181 |
+
return output_file, title
|
182 |
+
|
183 |
+
# Music2emo クラス(既存コード)
|
184 |
class Music2emo:
|
185 |
+
def __init__(self,
|
186 |
+
name="amaai-lab/music2emo",
|
187 |
+
device="cuda:0",
|
188 |
+
cache_dir=None,
|
189 |
+
local_files_only=False):
|
|
|
|
|
|
|
|
|
|
|
190 |
model_weights = "saved_models/J_all.ckpt"
|
191 |
self.device = device
|
|
|
192 |
self.feature_extractor = FeatureExtractorMERT(model_name='m-a-p/MERT-v1-95M', device=self.device, sr=resample_rate)
|
193 |
self.model_weights = model_weights
|
|
|
194 |
self.music2emo_model = FeedforwardModelMTAttnCK(
|
195 |
+
input_size=768 * 2,
|
196 |
output_size_classification=56,
|
197 |
output_size_regression=2
|
198 |
)
|
|
|
199 |
checkpoint = torch.load(self.model_weights, map_location=self.device, weights_only=False)
|
200 |
+
state_dict = {key.replace("model.", ""): value for key, value in checkpoint["state_dict"].items()}
|
|
|
|
|
|
|
|
|
|
|
201 |
model_keys = set(self.music2emo_model.state_dict().keys())
|
202 |
filtered_state_dict = {key: value for key, value in state_dict.items() if key in model_keys}
|
|
|
|
|
203 |
self.music2emo_model.load_state_dict(filtered_state_dict)
|
|
|
204 |
self.music2emo_model.to(self.device)
|
205 |
self.music2emo_model.eval()
|
|
|
206 |
self.config = HParams.load("./inference/data/run_config.yaml")
|
207 |
self.config.feature['large_voca'] = True
|
208 |
self.config.model['num_chords'] = 170
|
209 |
model_file = './inference/data/btc_model_large_voca.pt'
|
210 |
self.idx_to_voca = idx2voca_chord()
|
211 |
self.btc_model = BTC_model(config=self.config.model).to(self.device)
|
|
|
212 |
if os.path.isfile(model_file):
|
213 |
checkpoint = torch.load(model_file, map_location=self.device)
|
214 |
self.mean = checkpoint['mean']
|
215 |
self.std = checkpoint['std']
|
216 |
self.btc_model.load_state_dict(checkpoint['model'])
|
|
|
|
|
217 |
self.tonic_to_idx = {tonic: idx for idx, tonic in enumerate(tonic_signatures)}
|
218 |
self.mode_to_idx = {mode: idx for idx, mode in enumerate(mode_signatures)}
|
219 |
self.idx_to_tonic = {idx: tonic for tonic, idx in self.tonic_to_idx.items()}
|
220 |
self.idx_to_mode = {idx: mode for mode, idx in self.mode_to_idx.items()}
|
|
|
221 |
with open('inference/data/chord.json', 'r') as f:
|
222 |
self.chord_to_idx = json.load(f)
|
223 |
with open('inference/data/chord_inv.json', 'r') as f:
|
224 |
+
self.idx_to_chord = {int(k): v for k, v in json.load(f).items()}
|
|
|
225 |
with open('inference/data/chord_root.json') as json_file:
|
226 |
self.chordRootDic = json.load(json_file)
|
227 |
with open('inference/data/chord_attr.json') as json_file:
|
228 |
self.chordAttrDic = json.load(json_file)
|
229 |
|
230 |
+
def predict(self, audio, threshold=0.5):
|
|
|
|
|
|
|
231 |
feature_dir = Path("./inference/temp_out")
|
232 |
output_dir = Path("./inference/output")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
233 |
safe_remove_dir(feature_dir)
|
234 |
safe_remove_dir(output_dir)
|
|
|
235 |
feature_dir.mkdir(parents=True, exist_ok=True)
|
236 |
output_dir.mkdir(parents=True, exist_ok=True)
|
|
|
237 |
warnings.filterwarnings('ignore')
|
238 |
logger.logging_verbosity(1)
|
|
|
239 |
mert_dir = feature_dir / "mert"
|
240 |
mert_dir.mkdir(parents=True, exist_ok=True)
|
|
|
241 |
waveform, sample_rate = torchaudio.load(audio)
|
242 |
if waveform.shape[0] > 1:
|
243 |
waveform = waveform.mean(dim=0).unsqueeze(0)
|
244 |
waveform = waveform.squeeze()
|
245 |
waveform, sample_rate = resample_waveform(waveform, sample_rate, resample_rate)
|
246 |
+
if is_split:
|
|
|
247 |
segments = split_audio(waveform, sample_rate)
|
248 |
for i, segment in enumerate(segments):
|
249 |
segment_save_path = os.path.join(mert_dir, f"segment_{i}.npy")
|
|
|
251 |
else:
|
252 |
segment_save_path = os.path.join(mert_dir, f"segment_0.npy")
|
253 |
self.feature_extractor.extract_features_from_segment(waveform, sample_rate, segment_save_path)
|
|
|
|
|
|
|
254 |
segment_embeddings = []
|
255 |
+
layers_to_extract = [5,6]
|
256 |
+
for filename in sorted(os.listdir(mert_dir)):
|
257 |
file_path = os.path.join(mert_dir, filename)
|
258 |
if os.path.isfile(file_path) and filename.endswith('.npy'):
|
259 |
segment = np.load(file_path)
|
260 |
concatenated_features = np.concatenate(
|
261 |
[segment[:, layer_idx, :] for layer_idx in layers_to_extract], axis=1
|
262 |
)
|
263 |
+
concatenated_features = np.squeeze(concatenated_features)
|
264 |
segment_embeddings.append(concatenated_features)
|
|
|
265 |
segment_embeddings = np.array(segment_embeddings)
|
266 |
if len(segment_embeddings) > 0:
|
267 |
final_embedding_mert = np.mean(segment_embeddings, axis=0)
|
268 |
else:
|
269 |
final_embedding_mert = np.zeros((1536,))
|
270 |
+
final_embedding_mert = torch.from_numpy(final_embedding_mert).to(self.device)
|
|
|
|
|
|
|
|
|
|
|
271 |
audio_path = audio
|
272 |
+
audio_id = os.path.split(audio_path)[-1][:-4]
|
273 |
try:
|
274 |
feature, feature_per_second, song_length_second = audio_file_to_features(audio_path, self.config)
|
275 |
except:
|
276 |
+
logger.info("音声ファイルの読み込みに失敗しました : %s" % audio_path)
|
277 |
assert(False)
|
278 |
+
logger.info("音声ファイルの読み込みと特徴量計算に成功しました : %s" % audio_path)
|
|
|
|
|
279 |
feature = feature.T
|
280 |
feature = (feature - self.mean) / self.std
|
281 |
time_unit = feature_per_second
|
282 |
n_timestep = self.config.model['timestep']
|
|
|
283 |
num_pad = n_timestep - (feature.shape[0] % n_timestep)
|
284 |
feature = np.pad(feature, ((0, num_pad), (0, 0)), mode="constant", constant_values=0)
|
285 |
num_instance = feature.shape[0] // n_timestep
|
|
|
286 |
start_time = 0.0
|
287 |
lines = []
|
288 |
with torch.no_grad():
|
|
|
297 |
prev_chord = prediction[i].item()
|
298 |
continue
|
299 |
if prediction[i].item() != prev_chord:
|
300 |
+
lines.append('%.3f %.3f %s\n' % (start_time, time_unit * (n_timestep * t + i), self.idx_to_voca[prev_chord]))
|
|
|
301 |
start_time = time_unit * (n_timestep * t + i)
|
302 |
prev_chord = prediction[i].item()
|
303 |
if t == num_instance - 1 and i + num_pad == n_timestep:
|
304 |
if start_time != time_unit * (n_timestep * t + i):
|
305 |
lines.append('%.3f %.3f %s\n' % (start_time, time_unit * (n_timestep * t + i), self.idx_to_voca[prev_chord]))
|
306 |
break
|
|
|
307 |
save_path = os.path.join(feature_dir, os.path.split(audio_path)[-1].replace('.mp3', '').replace('.wav', '') + '.lab')
|
308 |
with open(save_path, 'w') as f:
|
309 |
for line in lines:
|
310 |
f.write(line)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
311 |
try:
|
312 |
midi_file = converter.parse(save_path.replace('.lab', '.midi'))
|
313 |
key_signature = str(midi_file.analyze('key'))
|
314 |
except Exception as e:
|
315 |
key_signature = "None"
|
|
|
316 |
key_parts = key_signature.split()
|
317 |
+
key_signature = sanitize_key_signature(key_parts[0])
|
318 |
key_type = key_parts[1] if len(key_parts) > 1 else 'major'
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
319 |
converted_lines = normalize_chord(save_path, key_signature, key_type)
|
|
|
320 |
lab_norm_path = save_path[:-4] + "_norm.lab"
|
|
|
|
|
321 |
with open(lab_norm_path, 'w') as f:
|
322 |
f.writelines(converted_lines)
|
|
|
323 |
chords = []
|
|
|
324 |
if not os.path.exists(lab_norm_path):
|
325 |
chords.append((float(0), float(0), "N"))
|
326 |
else:
|
|
|
328 |
for line in file:
|
329 |
start, end, chord = line.strip().split()
|
330 |
chords.append((float(start), float(end), chord))
|
|
|
331 |
encoded = []
|
332 |
+
encoded_root = []
|
333 |
+
encoded_attr = []
|
334 |
durations = []
|
|
|
335 |
for start, end, chord in chords:
|
336 |
chord_arr = chord.split(":")
|
337 |
if len(chord_arr) == 1:
|
338 |
chordRootID = self.chordRootDic[chord_arr[0]]
|
339 |
+
chordAttrID = 0 if chord_arr[0] in ["N", "X"] else 1
|
|
|
|
|
|
|
340 |
elif len(chord_arr) == 2:
|
341 |
chordRootID = self.chordRootDic[chord_arr[0]]
|
342 |
chordAttrID = self.chordAttrDic[chord_arr[1]]
|
343 |
encoded_root.append(chordRootID)
|
344 |
encoded_attr.append(chordAttrID)
|
|
|
345 |
if chord in self.chord_to_idx:
|
346 |
encoded.append(self.chord_to_idx[chord])
|
347 |
else:
|
348 |
+
print(f"警告: {chord} は chord.json に見つかりませんでした。スキップします。")
|
349 |
+
durations.append(end - start)
|
|
|
|
|
350 |
encoded_chords = np.array(encoded)
|
351 |
encoded_chords_root = np.array(encoded_root)
|
352 |
encoded_chords_attr = np.array(encoded_attr)
|
353 |
+
max_sequence_length = 100
|
|
|
|
|
|
|
|
|
354 |
if len(encoded_chords) > max_sequence_length:
|
|
|
355 |
encoded_chords = encoded_chords[:max_sequence_length]
|
356 |
encoded_chords_root = encoded_chords_root[:max_sequence_length]
|
357 |
encoded_chords_attr = encoded_chords_attr[:max_sequence_length]
|
|
|
358 |
else:
|
|
|
359 |
padding = [0] * (max_sequence_length - len(encoded_chords))
|
360 |
encoded_chords = np.concatenate([encoded_chords, padding])
|
361 |
encoded_chords_root = np.concatenate([encoded_chords_root, padding])
|
362 |
encoded_chords_attr = np.concatenate([encoded_chords_attr, padding])
|
|
|
|
|
363 |
chords_tensor = torch.tensor(encoded_chords, dtype=torch.long).to(self.device)
|
364 |
chords_root_tensor = torch.tensor(encoded_chords_root, dtype=torch.long).to(self.device)
|
365 |
chords_attr_tensor = torch.tensor(encoded_chords_attr, dtype=torch.long).to(self.device)
|
|
|
366 |
model_input_dic = {
|
367 |
"x_mert": final_embedding_mert.unsqueeze(0),
|
368 |
"x_chord": chords_tensor.unsqueeze(0),
|
369 |
"x_chord_root": chords_root_tensor.unsqueeze(0),
|
370 |
"x_chord_attr": chords_attr_tensor.unsqueeze(0),
|
371 |
+
"x_key": torch.tensor([self.mode_to_idx.get(key_type, 0)], dtype=torch.long).unsqueeze(0).to(self.device)
|
372 |
}
|
|
|
373 |
model_input_dic = {k: v.to(self.device) for k, v in model_input_dic.items()}
|
374 |
classification_output, regression_output = self.music2emo_model(model_input_dic)
|
375 |
+
tag_list = np.load("./inference/data/tag_list.npy")
|
|
|
|
|
376 |
tag_list = tag_list[127:]
|
377 |
mood_list = [t.replace("mood/theme---", "") for t in tag_list]
|
|
|
|
|
|
|
378 |
probs = torch.sigmoid(classification_output).squeeze().tolist()
|
|
|
|
|
379 |
predicted_moods_with_scores = [
|
380 |
+
{"mood": mood_list[i], "score": round(p, 4)}
|
381 |
for i, p in enumerate(probs) if p > threshold
|
382 |
]
|
|
|
|
|
383 |
predicted_moods_with_scores_all = [
|
384 |
+
{"mood": mood_list[i], "score": round(p, 4)}
|
385 |
for i, p in enumerate(probs)
|
386 |
]
|
|
|
|
|
|
|
387 |
predicted_moods_with_scores.sort(key=lambda x: x["score"], reverse=True)
|
|
|
388 |
valence, arousal = regression_output.squeeze().tolist()
|
|
|
389 |
model_output_dic = {
|
390 |
"valence": valence,
|
391 |
"arousal": arousal,
|
392 |
"predicted_moods": predicted_moods_with_scores,
|
393 |
"predicted_moods_all": predicted_moods_with_scores_all
|
394 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
395 |
return model_output_dic
|
396 |
|
397 |
+
# Music2Emo モデルの初期化
|
398 |
if torch.cuda.is_available():
|
399 |
music2emo = Music2emo()
|
400 |
else:
|
401 |
music2emo = Music2emo(device="cpu")
|
402 |
|
403 |
+
# 入力(音声ファイルまたはYouTube URL)を処理する関数
|
404 |
+
def process_input(audio, youtube_url, threshold):
|
405 |
+
if youtube_url and youtube_url.strip().startswith("http"):
|
406 |
+
# YouTube URL が入力されている場合、音声をダウンロード
|
407 |
+
audio_file, video_title = download_audio_from_youtube(youtube_url)
|
408 |
+
output_dic = music2emo.predict(audio_file, threshold)
|
409 |
+
output_text, va_chart, mood_chart = format_prediction(output_dic)
|
410 |
+
output_text += f"\n動画タイトル: {video_title}"
|
411 |
+
return output_text, va_chart, mood_chart
|
412 |
+
elif audio:
|
413 |
+
output_dic = music2emo.predict(audio, threshold)
|
414 |
+
return format_prediction(output_dic)
|
415 |
+
else:
|
416 |
+
return "音声ファイルまたは YouTube URL を入力してください。", None, None
|
417 |
+
|
418 |
+
# 解析結果のフォーマット関数
|
419 |
+
def format_prediction(model_output_dic):
|
420 |
+
valence = model_output_dic["valence"]
|
421 |
+
arousal = model_output_dic["arousal"]
|
422 |
+
predicted_moods_with_scores = model_output_dic["predicted_moods"]
|
423 |
+
predicted_moods_with_scores_all = model_output_dic["predicted_moods_all"]
|
424 |
+
va_chart = plot_valence_arousal(valence, arousal)
|
425 |
+
mood_chart = plot_mood_probabilities(predicted_moods_with_scores_all)
|
426 |
+
if predicted_moods_with_scores:
|
427 |
+
moods_text = ", ".join([f"{m['mood']} ({m['score']:.2f})" for m in predicted_moods_with_scores])
|
428 |
+
else:
|
429 |
+
moods_text = "顕著なムードは検出されませんでした。"
|
430 |
+
output_text = f"""🎭 ムードタグ: {moods_text}
|
431 |
+
|
432 |
+
💖 バレンス: {valence:.2f} (1〜9 スケール)
|
433 |
+
⚡ アラウザル: {arousal:.2f} (1〜9 スケール)"""
|
434 |
+
return output_text, va_chart, mood_chart
|
435 |
+
|
436 |
def plot_mood_probabilities(predicted_moods_with_scores):
|
|
|
437 |
if not predicted_moods_with_scores:
|
438 |
return None
|
|
|
|
|
439 |
moods = [m["mood"] for m in predicted_moods_with_scores]
|
440 |
probs = [m["score"] for m in predicted_moods_with_scores]
|
|
|
|
|
441 |
sorted_indices = np.argsort(probs)[::-1]
|
442 |
sorted_probs = [probs[i] for i in sorted_indices]
|
443 |
sorted_moods = [moods[i] for i in sorted_indices]
|
|
|
|
|
444 |
fig, ax = plt.subplots(figsize=(8, 4))
|
445 |
ax.barh(sorted_moods[:10], sorted_probs[:10], color="#4CAF50")
|
446 |
+
ax.set_xlabel("確率")
|
447 |
+
ax.set_title("上位10のムードタグ")
|
448 |
ax.invert_yaxis()
|
|
|
449 |
return fig
|
450 |
|
451 |
def plot_valence_arousal(valence, arousal):
|
|
|
452 |
fig, ax = plt.subplots(figsize=(4, 4))
|
453 |
ax.scatter(valence, arousal, color="red", s=100)
|
454 |
ax.set_xlim(1, 9)
|
455 |
ax.set_ylim(1, 9)
|
456 |
+
ax.axhline(y=5, color='gray', linestyle='--', linewidth=1)
|
457 |
+
ax.axvline(x=5, color='gray', linestyle='--', linewidth=1)
|
458 |
+
ax.set_xlabel("バレンス (ポジティブ度)")
|
459 |
+
ax.set_ylabel("アラウザル (活発度)")
|
460 |
+
ax.set_title("バレンス・アラウザル プロット")
|
|
|
|
|
|
|
|
|
|
|
461 |
ax.grid(True, linestyle="--", alpha=0.6)
|
|
|
462 |
return fig
|
463 |
|
464 |
+
# Gradio UI の設定
|
465 |
+
title = "🎵 Music2Emo:統一型音楽感情認識システム"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
466 |
description_text = """
|
467 |
+
<p>
|
468 |
+
音声ファイルまたは YouTube の URL を入力すると、Music2Emo が楽曲の感情的特徴を解析します。<br/><br/>
|
469 |
+
このデモでは、1) ムードタグ、2) バレンス(1〜9 スケール)、3) アラウザル(1〜9 スケール)を予測します。<br/><br/>
|
470 |
+
詳細は <a href="https://arxiv.org/abs/2502.03979" target="_blank">論文</a> をご参照ください。
|
471 |
</p>
|
472 |
"""
|
|
|
|
|
473 |
css = """
|
474 |
.gradio-container {
|
475 |
font-family: 'Inter', -apple-system, system-ui, sans-serif;
|
|
|
480 |
border-radius: 8px;
|
481 |
padding: 10px;
|
482 |
}
|
|
|
483 |
.gr-box {
|
484 |
padding-top: 25px !important;
|
485 |
}
|
|
|
488 |
with gr.Blocks(css=css) as demo:
|
489 |
gr.HTML(f"<h1 style='text-align: center;'>{title}</h1>")
|
490 |
gr.Markdown(description_text)
|
|
|
|
|
491 |
gr.Markdown("""
|
492 |
+
### 📝 注意事項:
|
493 |
+
- **対応音声フォーマット:** MP3, WAV
|
494 |
+
- **YouTube URL も入力可能です(任意)
|
495 |
+
- **推奨:** 高品質な音声ファイル
|
496 |
""")
|
|
|
497 |
with gr.Row():
|
|
|
498 |
with gr.Column(scale=1):
|
499 |
+
input_audio = gr.Audio(label="音声ファイルをアップロード", type="filepath")
|
500 |
+
youtube_url = gr.Textbox(label="YouTube URL (任意)", placeholder="例: https://youtu.be/XXXXXXX")
|
501 |
+
threshold = gr.Slider(minimum=0.0, maximum=1.0, value=0.5, step=0.01, label="ムード検出のしきい値", info="しきい値を調整してください")
|
502 |
+
predict_btn = gr.Button("🎭 感情解析を実行", variant="primary")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
503 |
with gr.Column(scale=1):
|
504 |
+
output_text = gr.Textbox(label="解析結果", lines=4, interactive=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
505 |
with gr.Row(equal_height=True):
|
506 |
+
mood_chart = gr.Plot(label="ムード確率", scale=2, elem_classes=["gr-box"])
|
507 |
+
va_chart = gr.Plot(label="バレンス・アラウザル", scale=1, elem_classes=["gr-box"])
|
|
|
508 |
predict_btn.click(
|
509 |
+
fn=process_input,
|
510 |
+
inputs=[input_audio, youtube_url, threshold],
|
511 |
outputs=[output_text, va_chart, mood_chart]
|
512 |
)
|
513 |
|
|
|
514 |
demo.queue().launch()
|
|
|
|
|
|