Trjtdtk

I am new to Machine Learning. So, for a project I am trying to classify instruments in .wav file. The dataset I am using is IRMAS.

Dataset contains 11 classes of instruments with recordings in 16 bit stereo wav format sampled at 44.1kHz of 3s for each instrument.

I am converting all audio files to spectrograms for CNN using

walk.py

import os
import sys
from spectrogram import convert_to_spectrogram as cts

current_path = sys.argv[1]
destination_path = sys.argv[2]

i = 1

for file in os.listdir(current_path):
 current_file = os.path.join(current_path, file)
 # cts(current_file, destination_path, os.path.splitext(file)[0]) # has problems with filenames having '.'
 cts(current_file, destination_path, str(i)) # if dont need same file name will start from 1 to number of files
 i += 1

spectrogram.py

import librosa
import librosa.display 
import numpy as np
import os

def convert_to_spectrogram(filepath, filedest, filename):

y, sr = librosa.load(filepath)
librosa.feature.melspectrogram(y=y, sr=sr)

D = np.abs(librosa.stft(y, hop_length = 300))**2
S = librosa.feature.melspectrogram(S=D)

# Passing through arguments to the Mel filters
S = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=256,
 fmax=8000)

import matplotlib.pyplot as plt
plt.figure(figsize=(10,4) , frameon=False)
librosa.display.specshow(librosa.power_to_db(S,
 ref=np.max),
 y_axis='mel', fmax=8000,
 x_axis='time')

plt.gca().set_axis_off()
plt.subplots_adjust(top = 1, bottom = 0, right = 1, left = 0, 
 hspace = 0, wspace = 0)
filename = filename + "1.jpg"
plt.savefig(os.path.join(filedest, filename))

This gives me spectrogram for all the audio files in the dataset.

Using this CNN model

from keras.models import Sequential
from keras.layers import Conv2D
from keras.layers import MaxPooling2D
from keras.layers import Flatten
from keras.layers import Dense
from keras.layers import Dropout
from keras.models import model_from_json
from keras.models import load_model

classifier = Sequential()

classifier.add(Conv2D(32, (3, 3), input_shape = (128, 128, 3), activation = 'relu'))
classifier.add(MaxPooling2D(pool_size = (2, 2)))
classifier.add(Dropout(0.5))
classifier.add(Conv2D(32, (3, 3), activation = 'relu'))
classifier.add(MaxPooling2D(pool_size = (2, 2)))
classifier.add(Dropout(0.5))
classifier.add(Flatten())
classifier.add(Dense(units = 32, activation = 'relu'))
classifier.add(Dense(units = 11, activation = 'softmax'))
classifier.compile(optimizer = 'adam', loss = 'categorical_crossentropy', metrics = ['accuracy'])

from keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(rescale = 1./255,)

test_datagen = ImageDataGenerator(rescale = 1./255)

training_set = train_datagen.flow_from_directory('dataset/training_set',
 target_size = (128, 128),
 batch_size = 128,
 class_mode = 'categorical')

test_set = test_datagen.flow_from_directory('dataset/test_set',
 target_size = (128, 128),
 batch_size = 64,
 class_mode = 'categorical')

classifier.fit_generator(training_set,
 steps_per_epoch = 1000,
 epochs = 10,
 validation_data = test_set,
 validation_steps = 500)

The accuracy I get is pretty good but val_acc is around 0.2 ~ 0.22

I want to improve the val_acc for this.
After searching on the net I got to know its overfitting and tried adding dropout but that didn't help.