def getSpecs(path):
specs = []
noise = []
# Get mel-specs for file
for spec in audio.specsFromFile(path,
rate=cfg.SAMPLE_RATE,
seconds=cfg.SPEC_LENGTH,
overlap=cfg.SPEC_OVERLAP,
minlen=cfg.SPEC_MINLEN,
fmin=cfg.SPEC_FMIN,
fmax=cfg.SPEC_FMAX,
spec_type=cfg.SPEC_TYPE,
shape=(cfg.IM_SIZE[1], cfg.IM_SIZE[0])):
# Determine signal to noise ratio
s2n = audio.signal2noise(spec)
specs.append(spec)
noise.append(s2n)
# Shuffle arrays (we want to select randomly later)
specs, noise = shuffle(specs, noise, random_state=RANDOM)
return specs, noise
def analyzeFile(soundscape, test_function):
ncnt = 0
# Store analysis here
analysis = {}
# Keep track of timestamps
pred_start = 0
# Set species list accordingly
setSpeciesList(cfg.DEPLOYMENT_LOCATION[0], cfg.DEPLOYMENT_LOCATION[1],
cfg.DEPLOYMENT_WEEK)
# Get specs for file
spec_batch = []
for spec in audio.specsFromFile(soundscape,
rate=cfg.SAMPLE_RATE,
seconds=cfg.SPEC_LENGTH,
overlap=cfg.SPEC_OVERLAP,
minlen=cfg.SPEC_MINLEN,
fmin=cfg.SPEC_FMIN,
fmax=cfg.SPEC_FMAX,
win_len=cfg.WIN_LEN,
spec_type=cfg.SPEC_TYPE,
magnitude_scale=cfg.MAGNITUDE_SCALE,
bandpass=True,
shape=(cfg.IM_SIZE[1], cfg.IM_SIZE[0]),
offset=0,
duration=None):
# Prepare as input
spec = model.prepareInput(spec)
# Add to batch
if len(spec_batch) > 0:
spec_batch = np.vstack((spec_batch, spec))
else:
spec_batch = spec
# Do we have enough specs for a prediction?
if len(spec_batch) >= cfg.SPECS_PER_PREDICTION:
# Make prediction
p, _ = model.predict(spec_batch, test_function)
# Calculate next timestamp
pred_end = pred_start + cfg.SPEC_LENGTH + (
(len(spec_batch) - 1) * (cfg.SPEC_LENGTH - cfg.SPEC_OVERLAP))
# Store prediction
analysis[getTimestamp(pred_start, pred_end)] = p
# Advance to next timestamp
pred_start = pred_end - cfg.SPEC_OVERLAP
spec_batch = []
return analysis
def getSpecBatches(split):
# Random Seed
random = cfg.getRandomState()
# Make predictions for every testfile
for t in split:
# Spec batch
spec_batch = []
# Get specs for file
for spec in audio.specsFromFile(t[0],
cfg.SAMPLE_RATE,
cfg.SPEC_LENGTH,
cfg.SPEC_OVERLAP,
cfg.SPEC_MINLEN,
shape=(cfg.IM_SIZE[1], cfg.IM_SIZE[0]),
fmin=cfg.SPEC_FMIN,
fmax=cfg.SPEC_FMAX,
spec_type=cfg.SPEC_TYPE):
# Resize spec
spec = image.resize(spec,
cfg.IM_SIZE[0],
cfg.IM_SIZE[1],
mode=cfg.RESIZE_MODE)
# Normalize spec
spec = image.normalize(spec, cfg.ZERO_CENTERED_NORMALIZATION)
# Prepare as input
spec = image.prepare(spec)
# Add to batch
if len(spec_batch) > 0:
spec_batch = np.vstack((spec_batch, spec))
else:
spec_batch = spec
# Batch too large?
if spec_batch.shape[0] >= cfg.MAX_SPECS_PER_FILE:
break
# No specs?
if len(spec_batch) == 0:
spec = random.normal(0.0, 1.0, (cfg.IM_SIZE[1], cfg.IM_SIZE[0]))
spec_batch = image.prepare(spec)
# Shuffle spec batch
spec_batch = shuffle(spec_batch, random_state=random)
# yield batch, labels and filename
yield spec_batch[:cfg.MAX_SPECS_PER_FILE], t[1], t[0].split(os.sep)[-1]
def getSpecs(path):
specs = []
noise = []
for spec in audio.specsFromFile(path,
rate=cfg.SAMPLE_RATE,
seconds=cfg.SPEC_LENGTH,
overlap=cfg.SPEC_OVERLAP,
minlen=cfg.SPEC_MINLEN,
fmin=cfg.SPEC_FMIN,
fmax=cfg.SPEC_FMAX,
spec_type=cfg.SPEC_TYPE,
shape=(cfg.IM_SIZE[1], cfg.IM_SIZE[0])):
s2n = audio.signal2noise(spec)
specs.append(spec)
noise.append(s2n)
specs, noise = shuffle(specs, noise, random_state=RANDOM)
return specs, noise
def getSpecBatches(split):
# Random Seed
random = cfg.getRandomState()
# Make predictions for every testfile
for t in split:
# Spec batch
spec_batch = []
# Keep track of timestamps
pred_start = 0
# Get specs for file
for spec in audio.specsFromFile(t[0],
cfg.SAMPLE_RATE,
cfg.SPEC_LENGTH,
cfg.SPEC_OVERLAP,
cfg.SPEC_MINLEN,
shape=(cfg.IM_SIZE[1], cfg.IM_SIZE[0]),
fmin=cfg.SPEC_FMIN,
fmax=cfg.SPEC_FMAX):
# Resize spec
spec = image.resize(spec,
cfg.IM_SIZE[0],
cfg.IM_SIZE[1],
mode=cfg.RESIZE_MODE)
# Normalize spec
spec = image.normalize(spec, cfg.ZERO_CENTERED_NORMALIZATION)
# Prepare as input
spec = image.prepare(spec)
# Add to batch
if len(spec_batch) > 0:
spec_batch = np.vstack((spec_batch, spec))
else:
spec_batch = spec
# Batch too large?
if spec_batch.shape[0] >= cfg.MAX_SPECS_PER_FILE:
break
# Do we have enough specs for a prediction?
if len(spec_batch) >= cfg.SPECS_PER_PREDICTION:
# Calculate next timestamp
pred_end = pred_start + cfg.SPEC_LENGTH + (
(len(spec_batch) - 1) *
(cfg.SPEC_LENGTH - cfg.SPEC_OVERLAP))
# Store prediction
ts = getTimestamp(int(pred_start), int(pred_end))
# Advance to next timestamp
pred_start = pred_end - cfg.SPEC_OVERLAP
yield spec_batch, t[1], ts, t[0].split(os.sep)[-1]
# Spec batch
spec_batch = []
def main():
parser = argparse.ArgumentParser()
parser.add_argument('--dataroot', default="", help='path to dataset')
#parser.add_argument('--model_path',default = "/media/labhdd/ASAS/V3/110modelv3.pt",help='path to model')
parser.add_argument('--model_path',
default="/media/labhdd/ASAS/Resnet34_2/",
help='path to model')
parser.add_argument('--csv_path',
default="./train_v3.csv",
help='path to label csv')
parser.add_argument('--cuda', action='store_true', help='enables cuda')
parser.add_argument('--ignore_prob', default=1e-4, type=float)
opt = parser.parse_args()
print(opt)
# # # # # # # # # # # # # # # # # # # # # # # # # # # # #
# You should read file here and call power spectrum to #
# calculate the value and make decision #
wav_files = [
f for f in sorted(os.listdir(cfg.TESTSET_PATH)) if f[0] != '.'
]
# # # # # # # # # # # # # # # # # # # # # # # # # # # # #
dirpath = 'resnet_34_2'
if not os.path.exists(dirpath):
os.makedirs(dirpath)
for inx in range(5, 120, 5): # using different epochs
model = model_prediction(opt, inx)
label = model.construct_label_dict()
# # # # # # # # # # # # # # # # # # # # # # # # # # # # #
#
filepath = os.path.join(dirpath,
'valset_s2n001_resnet34_%3d.csv' % (inx))
f = open(filepath, 'w+')
median_dict = load_median()
#result = []
#predict_class = []
#predict_prob = []
submission = []
SPEC_SIGNAL_THRESHOLD = 0.001
num = 0
predict_prob_threshold = 0.0
for file in wav_files:
if num == 5:
break
#spec_flux = select_frame_criterion(file)
#print(spec_flux,file = f)
# Get specs for file
cnt2 = 1
predict_class = []
predict_prob = []
predict_s2n = []
accept_spec_num = 0
for spec in audio.specsFromFile(
os.path.join(cfg.TESTSET_PATH, file),
cfg.SAMPLE_RATE,
cfg.SPEC_LENGTH,
cfg.SPEC_OVERLAP,
cfg.SPEC_MINLEN,
shape=(cfg.IM_SIZE[1], cfg.IM_SIZE[0]),
fmin=cfg.SPEC_FMIN,
fmax=cfg.SPEC_FMAX,
spec_type=cfg.SPEC_TYPE):
s2n = audio.signal2noise(spec)
#print(s2n,file = f)
# Above SIGNAL_THRESHOLD?
if s2n >= SPEC_SIGNAL_THRESHOLD:
# Resize spec
#spec = image.resize(spec, cfg.IM_SIZE[0], cfg.IM_SIZE[1], mode=cfg.RESIZE_MODE)
# Normalize spec
#spec = image.normalize(spec, cfg.ZERO_CENTERED_NORMALIZATION)
# Prepare as input
spec = image.prepare(spec)
k = np.random.rand(1, 1, 256, 128)
confidence, prob = model.get_probability(spec, opt)
# print(label)
probdict = {}
for i in range(659):
if prob[i] < opt.ignore_prob:
pass
else:
probdict[label[i]] = prob[i]
order = sorted(probdict.items(),
key=lambda x: x[1],
reverse=True)
#print(file + '_' + str(c) + ':' + 'confidence = ' + str(confidence)+'\n',file = f1)
#print(file + '_' + str(cnt2) + ':' + 'probablity = ' + str(order[0])+'\n',file = f2)
#print(order[0][1])
if (order[0][1] > predict_prob_threshold): # select all
start = cnt2 - cnt2 % 5
end = start + 5
timestamp = getTimestamp(start, end) #type:string
print(median_dict[file] + ';' + timestamp + ';' +
order[0][0] + ';' + str(1),
file=f)
submission.append([
median_dict[file] + ';' + timestamp + ';' +
order[0][0] + ';' + str(1)
])
accept_spec_num += 1
cnt2 = cnt2 + 1
print('number of audio file : ' + str(num + 1) + ', containing ' +
str(accept_spec_num) + ' specs')
num += 1
del model