search

MiteshPuthran / Speech-Emotion-Analyzer

The neural network model is capable of detecting five different male/female emotions from audio speeches. (Deep Learning, NLP, Python)
MIT License
1.3k stars 438 forks source link

TypeError: '<' not supported between instances of 'str' and 'int' #19

Closed uneverknwwhoim closed 5 years ago

uneverknwwhoim commented 5 years ago

from keras.utils import np_utils from sklearn.preprocessing import LabelEncoder

X_train = np.array(trainfeatures) y_train = np.array(trainlabel) X_test = np.array(testfeatures) y_test = np.array(testlabel)

lb = LabelEncoder()

y_train = np_utils.to_categorical(lb.fit_transform(y_train)) y_test = np_utils.to_categorical(lb.fit_transform(y_test))

ERROR AT THIS LINE... please help

uneverknwwhoim commented 5 years ago

can anyone help me on this?? please

MiteshPuthran commented 5 years ago

@uneverknwwhoim can you post the error you encounter and at which specific line are you getting it?

uneverknwwhoim commented 5 years ago

image

uneverknwwhoim commented 5 years ago

image

uneverknwwhoim commented 5 years ago

checkout the screenshot.. and pls help.. thank u in advance

MiteshPuthran commented 5 years ago

print your (y_train) and (y_test) and see if the labels are all in the array.

uneverknwwhoim commented 5 years ago

image it is listing out as follows

MiteshPuthran commented 5 years ago

There are issues with your labels in (y_test). You can see that there are integers like 0 which shouldn't exist.

uneverknwwhoim commented 5 years ago

How do i clear that anyidea? I didnt do anything i just loaded my audio data set and ran your project.. Please help me as i am a beginner.. i would be very thankful to u:)

MiteshPuthran commented 5 years ago

There might be either of two issues here. Either those labels have been converted to zero or those zeros are extra. Check this by checking the shape of the training and test dataset. Print shape(X_train), shape(y_train), shape(x_test) and shape(y_test). train and test shapes should match.

Nandy-Saran commented 5 years ago

Hi Mithesh! I am also getting the same error as uneverknwwhoim. I checked the dataset, it has 1851 files which is a combination of RAVDESS and SAVEE. While dividing data into testing and training sets, the size of the testing and training sets differ. Also X_train and y_train doesn't have the same shape.

I am getting NaN values which are then replaced by 0 after this code : rnewdf=rnewdf.fillna(0). Please do provide some assistance as soon as possible

Nandy-Saran commented 5 years ago

I will share you my folder of the audio files if needed.

MiteshPuthran commented 5 years ago

@Nandy-Saran You won't have same shape for X_train and y_train. They should have same number of rows but features will differ. It doesn't really matter how many number of files you have before splitting.

uneverknwwhoim commented 5 years ago

@Nandy-Saran did u solve the problem ??

uneverknwwhoim commented 5 years ago

@MITESHPUTHRANNEU rows are same though and features different when i checked.. but still the same error.. please help... image

uneverknwwhoim commented 5 years ago

@MITESHPUTHRANNEU i even tried to delete the NaN values using dropna() but then i am getting y_train and X_train features different..

uneverknwwhoim commented 5 years ago

@MITESHPUTHRANNEU i would be very glad if u help me out of this asap..

MiteshPuthran commented 5 years ago

@uneverknwwhoim The problem is as you stated. If you don't remove the null values, then you would be getting the error you stated at the top. If you remove the null values then the shape of X_train and y_train don't match. So ultimately the problem is 'y'. For some reason there are null labels in your dataset. You have make sure to fix that, then everything would be fine.

uneverknwwhoim commented 5 years ago

Can i get a dataset link of yours?

MiteshPuthran commented 5 years ago

@uneverknwwhoim I have linked the datasets in the readme file and those are only ones which I used for this project.

uneverknwwhoim commented 5 years ago

I am getting an accuracy of 0.0806 %

I have included 2k+ audio files still am not getting the accuracy of 70-80%? What to do?

MiteshPuthran commented 5 years ago

@uneverknwwhoim try augmenting the data and using different sampling rate to extract more features per audio file.

Nikhilbharadwaj08 commented 5 years ago

Hi Mithesh! I am also getting the same error as uneverknwwhoim. I checked the dataset, it has 1851 files which is a combination of RAVDESS and SAVEE. While dividing data into testing and training sets, the size of the testing and training sets differ. Also X_train and y_train doesn't have the same shape.

I am getting NaN values which are then replaced by 0 after this code : rnewdf=rnewdf.fillna(0). Please do provide some assistance as soon as possible

Hi Can you please share me the datasets!! I'm not able get those numbers!!