Speech Emotion Recogination

DataSet

import os
Root = "/content/drive/MyDrive/Colab_Notebooks/RAVDESS_Emotional_speech_audio"
os.chdir(Root)


ls

modelForPrediction1.sav modelForPrediction.sav speech-emotion-recognition-ravdess-data/ Speech_Emotion_Recognition_with_librosa.ipynb standardScalar.sav

import librosa
import soundfile
import os, glob, pickle
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.neural_network import MLPClassifier
from sklearn.metrics import accuracy_score



#Extract features (mfcc, chroma, mel) from a sound file
def extract_feature(file_name, mfcc, chroma, mel):
    with soundfile.SoundFile(file_name) as sound_file:
        X = sound_file.read(dtype="float32")
        sample_rate=sound_file.samplerate
        if chroma:
            stft=np.abs(librosa.stft(X))
        result=np.array([])
        if mfcc:
            mfccs=np.mean(librosa.feature.mfcc(y=X, sr=sample_rate, n_mfcc=40).T, axis=0)
            result=np.hstack((result, mfccs))
        if chroma:
            chroma=np.mean(librosa.feature.chroma_stft(S=stft, sr=sample_rate).T,axis=0)
            result=np.hstack((result, chroma))
        if mel:
            mel=np.mean(librosa.feature.melspectrogram(X, sr=sample_rate).T,axis=0)
            result=np.hstack((result, mel))
    return result


# Emotions in the RAVDESS dataset
emotions={
  '01':'neutral',
  '02':'calm',
  '03':'happy',
  '04':'sad',
  '05':'angry',
  '06':'fearful',
  '07':'disgust',
  '08':'surprised'
}

#Emotions to observe
observed_emotions=['calm', 'happy', 'fearful', 'disgust']


#Load the data and extract features for each sound file
def load_data(test_size=0.2):
    x,y=[],[]
    for file in glob.glob("/content/drive/MyDrive/Colab_Notebooks/RAVDESS_Emotional_speech_audio/speech-emotion-recognition-ravdess-data/Actor_*/*.wav"):
        file_name=os.path.basename(file)
        emotion=emotions[file_name.split("-")[2]]
        if emotion not in observed_emotions:
            continue
        feature=extract_feature(file, mfcc=True, chroma=True, mel=True)
        x.append(feature)
        y.append(emotion)
    return train_test_split(np.array(x), y, test_size=test_size, random_state=9)


#Split the dataset
x_train,x_test,y_train,y_test=load_data(test_size=0.25)

x_train

This is Output
array([[-6.02389954e+02, 5.97717743e+01, 8.60734844e+00, ..., 2.24425294e-05, 7.05290176e-06, 3.74911019e-06], [-6.64690369e+02, 6.82226181e+01, 6.91438007e+00, ..., 1.92348180e-05, 1.16888250e-05, 1.09572538e-05], [-5.56770630e+02, 3.49958611e+01, -1.21606884e+01, ..., 1.56850641e-04, 9.86818704e-05, 6.10335883e-05], ..., [-6.41358337e+02, 4.56047516e+01, 3.17263484e-01, ..., 3.32857708e-05, 2.42486913e-05, 1.74304023e-05], [-6.41742493e+02, 3.81749878e+01, -8.41347885e+00, ..., 3.26658337e-05, 2.97957540e-05, 2.17277611e-05], [-7.70246155e+02, 3.43720894e+01, 5.50091887e+00, ..., 4.58828936e-06, 2.15270302e-06, 1.44739533e-06]])

#Get the shape of the training and testing datasets
print((x_train.shape[0], x_test.shape[0]))


#Initialize the Multi Layer Perceptron Classifier
model=MLPClassifier(alpha=0.01, batch_size=256, epsilon=1e-08, hidden_layer_sizes=(300,), learning_rate='adaptive', max_iter=500)

#Train the model
model.fit(x_train,y_train)

Output
MLPClassifier(activation='relu', alpha=0.01, batch_size=256, beta_1=0.9, beta_2=0.999, early_stopping=False, epsilon=1e-08, hidden_layer_sizes=(300,), learning_rate='adaptive', learning_rate_init=0.001, max_fun=15000, max_iter=500, momentum=0.9, n_iter_no_change=10, nesterovs_momentum=True, power_t=0.5, random_state=None, shuffle=True, solver='adam', tol=0.0001, validation_fraction=0.1, verbose=False, warm_start=False)

#Predict for the test set
y_pred=model.predict(x_test)

y_pred
array(['calm', 'disgust', 'calm', 'happy', 'calm', 'happy', 'disgust', 'calm', 'happy', 'fearful', 'calm', 'fearful', 'disgust', 'fearful', 'fearful', 'calm', 'happy', 'fearful', 'disgust', 'calm', 'happy', 'happy', 'fearful', 'happy', 'happy', 'calm', 'disgust', 'calm', 'happy', 'happy', 'happy', 'calm', 'happy', 'happy', 'fearful', 'calm', 'disgust', 'calm', 'happy', 'disgust', 'calm', 'disgust', 'happy', 'happy', 'calm', 'fearful', 'calm', 'fearful', 'calm', 'happy', 'happy', 'disgust', 'fearful', 'calm', 'fearful', 'happy', 'fearful', 'disgust', 'disgust', 'calm', 'happy', 'fearful', 'disgust', 'fearful', 'fearful', 'calm', 'happy', 'happy', 'calm', 'happy', 'fearful', 'calm', 'calm', 'calm', 'disgust', 'calm', 'fearful', 'happy', 'happy', 'disgust', 'happy', 'fearful', 'fearful', 'fearful', 'calm', 'calm', 'calm', 'disgust', 'happy', 'calm', 'happy', 'calm', 'calm', 'fearful', 'calm', 'calm', 'calm', 'fearful', 'happy', 'fearful', 'calm', 'calm', 'calm', 'disgust', 'fearful', 'happy', 'disgust', 'happy', 'fearful', 'calm', 'happy', 'fearful', 'fearful', 'calm', 'happy', 'calm', 'disgust', 'calm', 'calm', 'disgust', 'calm', 'fearful', 'calm', 'calm', 'calm', 'disgust', 'fearful', 'calm', 'happy', 'fearful', 'calm', 'fearful', 'happy', 'fearful', 'calm', 'fearful', 'happy', 'happy', 'happy', 'fearful', 'disgust', 'fearful', 'disgust', 'calm', 'fearful', 'disgust', 'happy', 'disgust', 'disgust', 'calm', 'calm', 'happy', 'fearful', 'calm', 'fearful', 'calm', 'disgust', 'happy', 'happy', 'calm', 'disgust', 'calm', 'fearful', 'disgust', 'happy', 'calm', 'calm', 'calm', 'disgust', 'fearful', 'calm', 'happy', 'fearful', 'happy', 'calm', 'calm', 'fearful', 'disgust', 'happy', 'disgust', 'calm', 'calm', 'calm', 'disgust', 'disgust', 'calm', 'calm', 'fearful', 'happy', 'disgust', 'fearful', 'happy'], dtype='<U7')

#Calculate the accuracy of our model
accuracy=accuracy_score(y_true=y_test, y_pred=y_pred)

#Print the accuracy
print("Accuracy: {:.2f}%".format(accuracy*100))

from sklearn.metrics import accuracy_score, f1_score

f1_score(y_test, y_pred,average=None)

import pandas as pd
df=pd.DataFrame({'Actual': y_test, 'Predicted':y_pred})
df.head(20)

ActualPredicted
0calmcalm
1disgustdisgust
2calmcalm
3happyhappy
4happycalm
5happyhappy
6disgustdisgust
7disgustcalm
8happyhappy
9fearfulfearful
10calmcalm
11disgustfearful
12disgustdisgust
13fearfulfearful
14disgustfearful
15calmcalm
16happyhappy
17fearfulfearful
18disgustdisgust
19calmcalm

import pickle
# Writing different model files to file
with open( 'modelForPrediction1.sav', 'wb') as f:
    pickle.dump(model,f)

filename = 'modelForPrediction1.sav'
loaded_model = pickle.load(open(filename, 'rb')) # loading the model file from the storage

feature=extract_feature("/content/drive/MyDrive/Colab_Notebooks/RAVDESS_Emotional_speech_audio/speech-emotion-recognition-ravdess-data/Actor_01/03-01-01-01-01-01-01.wav", mfcc=True, chroma=True, mel=True)

feature=feature.reshape(1,-1)

prediction=loaded_model.predict(feature)
prediction

Here its Ennd

Comments

Post a Comment

Popular posts from this blog