def test_generator_fake_loader():
dataset_path = 'data/UrbanSound8K/'
urbansound8k.default_path = dataset_path
data = urbansound8k.load_dataset()
folds, test = urbansound8k.folds(data)
data_length = 16
batch_size = 8
frames = 72
bands = 32
n_classes = 10
def zero_loader(s):
#assert
return numpy.zeros((bands, frames, 1))
fold = folds[0][0]
X = fold[0:data_length]
Y = fold.classID[0:data_length]
g = train.dataframe_generator(X,
Y,
loader=zero_loader,
batchsize=batch_size,
n_classes=n_classes)
n_batches = 3
batches = list(itertools.islice(g, n_batches))
assert len(batches) == n_batches
assert len(batches[0]) == 2 # X,y
assert batches[0][0].shape == (batch_size, bands, frames, 1)
assert batches[0][1].shape == (batch_size, n_classes)
def test_precompute():
settings = dict(
feature='mels',
samplerate=16000,
n_mels=32,
fmin=0,
fmax=8000,
n_fft=512,
hop_length=256,
augmentations=12,
)
dir = './pre2'
if os.path.exists(dir):
shutil.rmtree(dir)
workdir = os.path.abspath(os.path.join(os.path.dirname(__file__), '../data/'))
data = urbansound8k.load_dataset()
urbansound8k.maybe_download_dataset(workdir)
d = os.path.join(dir, features.settings_id(settings))
expect_path = features.feature_path(data.iloc[0], d)
assert not os.path.exists(expect_path), expect_path
preprocess.precompute(data[0:4], settings, out_dir=d, verbose=0, force=True, n_jobs=2)
assert os.path.exists(expect_path), expect_path
def loadfiles():
dataset = urbansound8k.load_dataset()
dataset['embeddings'] = None
# """print(dataset)
for index, sample in dataset.iterrows():
path = embedding_path(sample,"/Users/giovannibirkelund/Documents/GitHub/ESC-CNN-microcontroller/data/embeddedSounds")
data = np.load(path)
emb = data['embedding']
dataset.at[index,'embeddings'] = emb
return dataset
def embedFiles(datasetPath, newDir):
soundfiles = urbansound8k.load_dataset()
model = openl3.models.load_embedding_model(input_repr="mel256", content_type="env",
embedding_size=512)
for index in range(len(soundfiles)):
sample = soundfiles.iloc[index, :] #fetch each sample from the dataset
path = urbansound8k.sample_path(sample,datasetPath) #find the path given the sample, and the directory the files lay in.
fold = sample["fold"] #get the fold for the sound from the sample
differentDir = newDir+str(fold) #assign directory to save file in by adding the respective fold to the path
openl3.process_file(path, suffix=None, output_dir=differentDir,model=model, content_type="env", embedding_size=512, hop_size=0.5, verbose=True) #save the embedded file in its respective fold in the differentDir directory
def test_folds():
data = urbansound8k.load_dataset()
f = urbansound8k.folds(data)
assert len(f) == 10
import sys
from microesc import urbansound8k
import pandas
import numpy
data = urbansound8k.load_dataset()
by_class = data.groupby('class')
foreground_ratio = by_class.apply(lambda r: numpy.mean(r['salience'] == 1))
table = pandas.DataFrame({
'Samples': by_class.count()['classID'],
'Duration (avg)': by_class.apply(lambda r: '%.2f s' % (r.end-r.start).mean()),
'In foreground': [ "{} %".format(int(100*r)) for r in foreground_ratio ]
})
out = table.to_latex(header=True, index=True, column_format="lrrr")
print(out)
outpath = sys.argv[1]
with open(outpath, 'w') as f:
f.write(out)