def read_data_to_array(path, max_len=21, rescale=False):
X = []
y = []
for wavfile in os.listdir(path):
if wavfile.startswith('.') or not wavfile.endswith('.wav'):
continue
y.append(LABEL_MAPPING[wavfile[5]])
X.append(wav2mfcc(os.path.join(path, wavfile), max_len, rescale))
y_counter = Counter(y)
print(y_counter.most_common())
label_to_int_dict = {
l: idx
for idx, (l, _) in enumerate(y_counter.most_common())
}
y = [label_to_int_dict[l] for l in y]
X, y = np.stack(X), np.stack(y)
y = to_categorical(y)
return X, y, label_to_int_dict
def predict(name):
data = {"path": name}
params = flask.request.json
if (params == None):
params = flask.request.args
# if parameters are found, return a prediction
if (params != None):
with graph.as_default():
sample = preprocess.wav2mfcc('C://Users//Stage//Downloads//' +
name + '.wav')
print(name)
sample_reshaped = sample.reshape(1, 40, 47, 1)
data["prediction"] = preprocess.get_labels()[0][np.argmax(
model.predict(sample_reshaped))]
data["success"] = True
# return a response in json format
return flask.jsonify(data)
def predictTest(name):
data = {"path": name}
params = flask.request.json
if (params == None):
params = flask.request.args
# if parameters are found, return a prediction
if (params != None):
with graph.as_default():
dir = "C://Users//Stage//final project//test//" + name
filename = random.choice(os.listdir(dir))
print(filename)
sample = preprocess.wav2mfcc(dir + "//" + filename)
print(name)
sample_reshaped = sample.reshape(1, 40, 47, 1)
data["prediction"] = preprocess.get_labels()[0][np.argmax(
model.predict(sample_reshaped))]
data["success"] = True
# return a response in json format
return flask.jsonify(data)
This script is used to make inferences from a trained Convolutional Neural Network on new data. The
data is given as arguments to the program. The Preprocess module is used to generate the MFCCs of the audio data and load them into
memory.
It requires keras to be installed.
'''
from preprocess import wav2mfcc
from keras.models import load_model
import numpy as np
import sys
import time
def predict(model, mfcc):
reshaped = mfcc.reshape(1, 20, 11, 1)
return model.predict(reshaped)[0]
PATH_TO_MODEL = 'trained.h5'
LABELS = ['car_horn', 'dog_bark']
initial = time.time()
model = load_model(PATH_TO_MODEL)
print(f'Model took {time.time() - initial} seconds to load.')
for path in sys.argv[1:]:
initial = time.time()
prediction = predict(model, wav2mfcc(path))
print(
f'Prediction for {path}: {LABELS[np.argmax(prediction)]}; {prediction}; prediction took {time.time() - initial} seconds.'
)
def predict(filepath, model=None): # predict english word based CNN
sample = wav2mfcc(filepath)
feature_dim_1, feature_dim_2, channel = 20, 11, 1
sample_reshaped = sample.reshape(1, feature_dim_1, feature_dim_2, channel)
return get_labels()[0][np.argmax(model.predict(sample_reshaped))]
def predict(filepath, model):
sample = wav2mfcc(filepath)
sample_reshaped = sample.reshape(1, feature_dim_1, feature_dim_2, channel)
return get_labels()[0][np.argmax(model.predict(sample_reshaped))]