Spaces:
Running
Running
admin
commited on
Commit
·
6bbbe9a
1
Parent(s):
7b7e565
merge try
Browse files
app.py
CHANGED
@@ -251,102 +251,87 @@ def circular_padding(spec: np.ndarray, end: int):
|
|
251 |
|
252 |
|
253 |
def wav2mel(audio_path: str, width=2, top_db=40):
|
254 |
-
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
|
259 |
-
|
260 |
-
|
261 |
-
count = total_frames // (width * sr) + 1
|
262 |
-
y = circular_padding(y, count * width * sr)
|
263 |
-
|
264 |
-
mel_spec = librosa.feature.melspectrogram(y=y, sr=sr)
|
265 |
-
log_mel_spec = librosa.power_to_db(mel_spec, ref=np.max)
|
266 |
-
dur = librosa.get_duration(y=y, sr=sr)
|
267 |
-
total_frames = log_mel_spec.shape[1]
|
268 |
-
step = int(width * total_frames / dur)
|
269 |
-
count = int(total_frames / step)
|
270 |
-
begin = int(0.5 * (total_frames - count * step))
|
271 |
-
end = begin + step * count
|
272 |
-
for i in range(begin, end, step):
|
273 |
-
librosa.display.specshow(log_mel_spec[:, i : i + step])
|
274 |
-
plt.axis("off")
|
275 |
-
plt.savefig(
|
276 |
-
f"{TEMP_DIR}/{i}.jpg",
|
277 |
-
bbox_inches="tight",
|
278 |
-
pad_inches=0.0,
|
279 |
-
)
|
280 |
-
plt.close()
|
281 |
|
282 |
-
|
283 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
284 |
|
285 |
|
286 |
def wav2cqt(audio_path: str, width=2, top_db=40):
|
287 |
-
|
288 |
-
|
289 |
-
|
290 |
-
|
291 |
-
|
292 |
-
|
293 |
-
|
294 |
-
count = total_frames // (width * sr) + 1
|
295 |
-
y = circular_padding(y, count * width * sr)
|
296 |
-
|
297 |
-
cqt_spec = librosa.cqt(y=y, sr=sr)
|
298 |
-
log_cqt_spec = librosa.power_to_db(np.abs(cqt_spec) ** 2, ref=np.max)
|
299 |
-
dur = librosa.get_duration(y=y, sr=sr)
|
300 |
-
total_frames = log_cqt_spec.shape[1]
|
301 |
-
step = int(width * total_frames / dur)
|
302 |
-
count = int(total_frames / step)
|
303 |
-
begin = int(0.5 * (total_frames - count * step))
|
304 |
-
end = begin + step * count
|
305 |
-
for i in range(begin, end, step):
|
306 |
-
librosa.display.specshow(log_cqt_spec[:, i : i + step])
|
307 |
-
plt.axis("off")
|
308 |
-
plt.savefig(
|
309 |
-
f"{TEMP_DIR}/{i}.jpg",
|
310 |
-
bbox_inches="tight",
|
311 |
-
pad_inches=0.0,
|
312 |
-
)
|
313 |
-
plt.close()
|
314 |
|
315 |
-
|
316 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
317 |
|
318 |
|
319 |
def wav2chroma(audio_path: str, width=2, top_db=40):
|
320 |
-
|
321 |
-
|
322 |
-
|
323 |
-
|
324 |
-
|
325 |
-
|
326 |
-
|
327 |
-
count = total_frames // (width * sr) + 1
|
328 |
-
y = circular_padding(y, count * width * sr)
|
329 |
-
|
330 |
-
chroma_spec = librosa.feature.chroma_stft(y=y, sr=sr)
|
331 |
-
log_chroma_spec = librosa.power_to_db(np.abs(chroma_spec) ** 2, ref=np.max)
|
332 |
-
dur = librosa.get_duration(y=y, sr=sr)
|
333 |
-
total_frames = log_chroma_spec.shape[1]
|
334 |
-
step = int(width * total_frames / dur)
|
335 |
-
count = int(total_frames / step)
|
336 |
-
begin = int(0.5 * (total_frames - count * step))
|
337 |
-
end = begin + step * count
|
338 |
-
for i in range(begin, end, step):
|
339 |
-
librosa.display.specshow(log_chroma_spec[:, i : i + step])
|
340 |
-
plt.axis("off")
|
341 |
-
plt.savefig(
|
342 |
-
f"{TEMP_DIR}/{i}.jpg",
|
343 |
-
bbox_inches="tight",
|
344 |
-
pad_inches=0.0,
|
345 |
-
)
|
346 |
-
plt.close()
|
347 |
|
348 |
-
|
349 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
350 |
|
351 |
|
352 |
def most_frequent_value(lst: list):
|
@@ -366,13 +351,15 @@ def infer(wav_path: str, log_name: str, folder_path=TEMP_DIR):
|
|
366 |
if not wav_path:
|
367 |
return None, "Please input an audio!"
|
368 |
|
|
|
|
|
369 |
try:
|
370 |
model = EvalNet(log_name, len(TRANSLATE)).model
|
|
|
|
|
371 |
except Exception as e:
|
372 |
return None, f"{e}"
|
373 |
|
374 |
-
spec = log_name.split("_")[-3]
|
375 |
-
eval("wav2%s" % spec)(wav_path)
|
376 |
jpgs = find_files(folder_path, ".jpg")
|
377 |
preds = []
|
378 |
for jpg in jpgs:
|
|
|
251 |
|
252 |
|
253 |
def wav2mel(audio_path: str, width=2, top_db=40):
|
254 |
+
y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
|
255 |
+
non_silents = librosa.effects.split(y, top_db=top_db)
|
256 |
+
y = np.concatenate([y[start:end] for start, end in non_silents])
|
257 |
+
total_frames = len(y)
|
258 |
+
if total_frames % (width * sr) != 0:
|
259 |
+
count = total_frames // (width * sr) + 1
|
260 |
+
y = circular_padding(y, count * width * sr)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
261 |
|
262 |
+
mel_spec = librosa.feature.melspectrogram(y=y, sr=sr)
|
263 |
+
log_mel_spec = librosa.power_to_db(mel_spec, ref=np.max)
|
264 |
+
dur = librosa.get_duration(y=y, sr=sr)
|
265 |
+
total_frames = log_mel_spec.shape[1]
|
266 |
+
step = int(width * total_frames / dur)
|
267 |
+
count = int(total_frames / step)
|
268 |
+
begin = int(0.5 * (total_frames - count * step))
|
269 |
+
end = begin + step * count
|
270 |
+
for i in range(begin, end, step):
|
271 |
+
librosa.display.specshow(log_mel_spec[:, i : i + step])
|
272 |
+
plt.axis("off")
|
273 |
+
plt.savefig(
|
274 |
+
f"{TEMP_DIR}/{i}.jpg",
|
275 |
+
bbox_inches="tight",
|
276 |
+
pad_inches=0.0,
|
277 |
+
)
|
278 |
+
plt.close()
|
279 |
|
280 |
|
281 |
def wav2cqt(audio_path: str, width=2, top_db=40):
|
282 |
+
y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
|
283 |
+
non_silents = librosa.effects.split(y, top_db=top_db)
|
284 |
+
y = np.concatenate([y[start:end] for start, end in non_silents])
|
285 |
+
total_frames = len(y)
|
286 |
+
if total_frames % (width * sr) != 0:
|
287 |
+
count = total_frames // (width * sr) + 1
|
288 |
+
y = circular_padding(y, count * width * sr)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
289 |
|
290 |
+
cqt_spec = librosa.cqt(y=y, sr=sr)
|
291 |
+
log_cqt_spec = librosa.power_to_db(np.abs(cqt_spec) ** 2, ref=np.max)
|
292 |
+
dur = librosa.get_duration(y=y, sr=sr)
|
293 |
+
total_frames = log_cqt_spec.shape[1]
|
294 |
+
step = int(width * total_frames / dur)
|
295 |
+
count = int(total_frames / step)
|
296 |
+
begin = int(0.5 * (total_frames - count * step))
|
297 |
+
end = begin + step * count
|
298 |
+
for i in range(begin, end, step):
|
299 |
+
librosa.display.specshow(log_cqt_spec[:, i : i + step])
|
300 |
+
plt.axis("off")
|
301 |
+
plt.savefig(
|
302 |
+
f"{TEMP_DIR}/{i}.jpg",
|
303 |
+
bbox_inches="tight",
|
304 |
+
pad_inches=0.0,
|
305 |
+
)
|
306 |
+
plt.close()
|
307 |
|
308 |
|
309 |
def wav2chroma(audio_path: str, width=2, top_db=40):
|
310 |
+
y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
|
311 |
+
non_silents = librosa.effects.split(y, top_db=top_db)
|
312 |
+
y = np.concatenate([y[start:end] for start, end in non_silents])
|
313 |
+
total_frames = len(y)
|
314 |
+
if total_frames % (width * sr) != 0:
|
315 |
+
count = total_frames // (width * sr) + 1
|
316 |
+
y = circular_padding(y, count * width * sr)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
317 |
|
318 |
+
chroma_spec = librosa.feature.chroma_stft(y=y, sr=sr)
|
319 |
+
log_chroma_spec = librosa.power_to_db(np.abs(chroma_spec) ** 2, ref=np.max)
|
320 |
+
dur = librosa.get_duration(y=y, sr=sr)
|
321 |
+
total_frames = log_chroma_spec.shape[1]
|
322 |
+
step = int(width * total_frames / dur)
|
323 |
+
count = int(total_frames / step)
|
324 |
+
begin = int(0.5 * (total_frames - count * step))
|
325 |
+
end = begin + step * count
|
326 |
+
for i in range(begin, end, step):
|
327 |
+
librosa.display.specshow(log_chroma_spec[:, i : i + step])
|
328 |
+
plt.axis("off")
|
329 |
+
plt.savefig(
|
330 |
+
f"{TEMP_DIR}/{i}.jpg",
|
331 |
+
bbox_inches="tight",
|
332 |
+
pad_inches=0.0,
|
333 |
+
)
|
334 |
+
plt.close()
|
335 |
|
336 |
|
337 |
def most_frequent_value(lst: list):
|
|
|
351 |
if not wav_path:
|
352 |
return None, "Please input an audio!"
|
353 |
|
354 |
+
spec = log_name.split("_")[-3]
|
355 |
+
os.makedirs(folder_path, exist_ok=True)
|
356 |
try:
|
357 |
model = EvalNet(log_name, len(TRANSLATE)).model
|
358 |
+
eval("wav2%s" % spec)(wav_path)
|
359 |
+
|
360 |
except Exception as e:
|
361 |
return None, f"{e}"
|
362 |
|
|
|
|
|
363 |
jpgs = find_files(folder_path, ".jpg")
|
364 |
preds = []
|
365 |
for jpg in jpgs:
|