#### Dependencies ######## Import Cometfor experiment tracking and visual tools
from comet_ml import Experiment
####import IPython.display as ipd
import numpy as np
import pandas as pd
import librosa
import matplotlib.pyplot as plt
from scipy.io import wavfile as wavfrom sklearn import metrics
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import train_test_split from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation
from keras.optimizers import Adam
from keras.utils import to_categorical

experiment = Experiment(api_key="API_KEY" ,project_name= "urbansound8k" )

# Load dataset
df = pd.read_csv('UrbanSound8K/metadata/UrbanSound8K.csv' )# Create a list of the class labels
labels  = list(df[ 'class' ].unique())# Let 's grab a single audio file from each class
files = dict()
for i in range(len(labels)):tmp = df[df[' class ']  == labels[i]][: 1 ].reset_index()path = 'UrbanSound8K/audio/fold{}/{}' .format(tmp[ 'fold' ][ 0 ], tmp[ 'slice_file_name' ][ 0 ])files[labels[i]] = path

fig = plt.figure(figsize=(15 , 15 ))
fig.subplots_adjust(hspace= 0.4 , wspace= 0.4 )
for i, label in enumerate(labels):fn = files[label]fig.add_subplot( 5 , 2 , i+ 1 )plt.title(label)data, sample_rate = librosa.load(fn)librosa.display.waveplot(data, sr= sample_rate)
plt.savefig( 'class_examples.png' )

# Log graphicof waveforms to Comet
experiment.log_image( 'class_examples.png' )

# Log audio files to Cometfor debugging
for label in labels:fn = files[label]experiment.log_audio(fn, metadata = { 'name' : label})

fn ='UrbanSound8K/audio/fold1/191431-9-0-66.wav'
librosa_audio, librosa_sample_rate = librosa.load(fn)
scipy_sample_rate, scipy_audio = wav.read(fn)print( "Original sample rate: {}" .format(scipy_sample_rate))
print( "Librosa sample rate: {}" .format(librosa_sample_rate))

print('Original audio file min~max range: {} to {}' .format(np.min(scipy_audio), np.max(scipy_audio)))
print( 'Librosa audio file min~max range: {0:.2f} to {0:.2f}' .format(np.min(librosa_audio), np.max(librosa_audio)))

plt.figure(figsize=(12 , 4 ))
plt.plot(scipy_audio)
plt.savefig( 'original_audio.png' )
experiment.log_image( 'original_audio.png' )

# Librosa: mono track
plt.figure(figsize=(12 , 4 ))
plt.plot(librosa_audio)
plt.savefig( 'librosa_audio.png' )
experiment.log_image( 'librosa_audio.png' )

mfccs = librosa.feature.mfcc(y=librosa_audio, sr=librosa_sample_rate, n_mfcc =40 )

print (mfccs.shape)

plt.figure(figsize=(8 , 8 ))
librosa.display.specshow(mfccs, sr=librosa_sample_rate, x_axis= 'time' )
plt.savefig( 'MFCCs.png' )
experiment.log_image( 'MFCCs.png' )

def extract_features(file_name):audio, sample_rate = librosa.load(file_name, res_type='kaiser_fast' ) mfccs = librosa.feature.mfcc(y=audio, sr=sample_rate, n_mfcc= 40 )mfccs_processed = np.mean(mfccs.T,axis= 0 )return mfccs_processed

features = []# Iterate through each sound file and extract the featuresfor index, row in metadata.iterrows():file_name = os.path.join(os.path.abspath(fulldatasetpath), 'fold' +str(row[ "fold" ])+ '/' ,str(row[ "slice_file_name" ]))class_label = row[ "class" ]data = extract_features(file_name)features.append([data, class_label])# Convert into a Panda dataframe
featuresdf = pd.DataFrame(features, columns=[ 'feature' , 'class_label' ])

featuresdf.head()

featuresdf.iloc[0 ][ 'feature' ]

array([-2 .1579300e +02 , 7.1666122e +01 ,
-1 .3181377e +02 , -5 .2091331e +01 ,
-2 .2115969e +01 , -2 .1764181e +01 ,
-1 .1183747e +01 , 1.8912683e +01 ,
6.7266388e +00 , 1.4556893e +01 ,
-1 .1782045e +01 , 2.3010368e +00 ,
-1 .7251305e +01 , 1.0052421e +01 ,
-6 .0095000e +00 , -1 .3153191e +00 ,
-1 .7693510e +01 , 1.1171228e +00 ,
-4 .3699470e +00 , 7.2629538e +00 ,
-1 .1815971e +01 , -7 .4952612e +00 ,
5.4577131e +00 , -2 .9442446e +00 ,
-5 .8693886e +00 , -9 .8654032e -02 ,
-3 .2121708e +00 , 4.6092505e +00 ,
-5 .8293257e +00 , -5 .3475075e +00 ,
1.3341187e +00 , 7.1307826e +00 ,
-7 .9450034e -02 , 1.7109241e +00 ,
-5 .6942000e +00 , -2 .9041715e +00 ,
3.0366952e +00 , -1 .6827590e +00 ,
-8 .8585770e -01 , 3.5438776e -01 ],
dtype=float32)

from sklearn.preprocessing import LabelEncoder
from keras.utils import to_categorical# Convert features and corresponding classification labels into numpy arrays
X = np.array(featuresdf.feature.tolist())
y = np.array(featuresdf.class_label.tolist())# Encode the classification labels
le = LabelEncoder()
yy = to_categorical(le.fit_transform(y))

# split the datasetfrom sklearn.model_selection import train_test_split x_train, x_test, y_train, y_test = train_test_split(X, yy, test_size= 0.2 , random_state = 127 )

num_labels = yy.shape[1 ]
filter_size = 2def build_model_graph(input_shape=( 40 ,)):model = Sequential()model.add(Dense( 256 ))model.add(Activation( 'relu' ))model.add(Dropout( 0.5 ))model.add(Dense( 256 ))model.add(Activation( 'relu' ))model.add(Dropout( 0.5 ))model.add(Dense(num_labels))model.add(Activation( 'softmax' ))# Compile the modelmodel.compile(loss= 'categorical_crossentropy' , metrics=[ 'accuracy' ], optimizer= 'adam' )return modelmodel = build_model_graph()

# Display model architecture summary
model.summary()# Calculate pre-training accuracy
score = model.evaluate(x_test, y_test, verbose=0 )
accuracy = 100 *score[ 1 ]

print("Pre-training accuracy: %.4f%%" % accuracy)

from keras.callbacks import ModelCheckpoint
from datetime import datetime num_epochs = 100
num_batch_size = 32model.fit(x_train, y_train, batch_size=num_batch_size, epochs=num_epochs, validation_data=(x_test, y_test), verbose= 1 )

# Evaluating the model on the training and testing set
score = model.evaluate(x_train, y_train, verbose=0 )
print( "Training Accuracy: {0:.2%}" .format(score[ 1 ]))score = model.evaluate(x_test, y_test, verbose= 0 )
print( "Testing Accuracy: {0:.2%}" .format(score[ 1 ]))