Dataset has no attribute 'data' in find_best_model in line 190

[AdamarisAinsley]

Dataset is not loaded into the file, when i try another speech dataset (SAVEE/CASIA.IEMOCAP), help me to know ,how to solve this issue?

[WakeUpYoimiya]

Dataset is not loaded into the file, when i try another speech dataset (SAVEE/CASIA.IEMOCAP), help me to know ,how to solve this issue?

Were there any error messages? I just solved these data problems and ran the program successfully, so maybe i can help u with these stuff.

[AdamarisAinsley]

It does not give any error but it gives 100% training accuracy, but reported accuracy on IEMOCAP is 82%...... [image: image.png]

[WakeUpYoimiya]

Have u modified the dataset.py? I think the validation in find_best_model.py is based on the file name of the data. Would you post the name of the data u used here as a sample?

[AdamarisAinsley]

I made changes in dataset.py file as follows:
import os
import subprocess as sp
import itertools
import librosa
import glob
import wave
import numpy as np
import python_speech_features as ps
import cPickle
import csv

class Dataset:

def __init__(self, path, dataset, decode=False):
    self.dataset = dataset
    if dataset == "IEMOCAP":
        self.classes = {0: 'ang', 1: 'sad', 2: 'hap', 3: 'neu', 4: 'fea', 5: 'fru', 6:'exc' }
        self.get_IEMOCAP_dataset(path)
def get_IEMOCAP_dataset(self, path):
    males = ['Ses01M','Ses02M','Ses03M','Ses04M','Ses05M']
    females = ['Ses01F','Ses02F','Ses03F','Ses04F','Ses05F']
    try:
        classes = {v: k for k, v in self.classes.iteritems()}
    except AttributeError:
        classes = {v: k for k, v in self.classes.items()}
    self.targets = []
    self.data = []
    self.train_sets = []
    self.test_sets = []
    get_data = True
    for speak_test in itertools.product(males, females):
        i = 0
        train = []
        test = []
        for audio in os.listdir(path):
         if(audio[0] == 'S'):
            sub_dir = os.path.join(path,audio,'sentences/wav')
            emoevl = os.path.join(path,audio,'dialog/EmoEvaluation')
            for sess in os.listdir(sub_dir):
                if(sess[7] == 'i'):
                    emotdir = emoevl+'/'+sess+'.txt'
                    emot_map = {}
                    with open(emotdir,'r') as emot_to_read:
                        while True:
                            line = emot_to_read.readline()
                            if not line:
                                break
                            if(line[0] == '['):
                                t = line.split()
                                emot_map[t[3]] = t[4]

                    file_dir = os.path.join(sub_dir, sess, '*.wav')
                    files = glob.glob(file_dir)
                    for filename in files:
                         wavname = filename.split("/")[-1][:-4]
                         f = open("Audio3.csv","a")
                         f.write(wavname)
                         f.write("\n")
                         emotion = emot_map[wavname]
                         f = open("features3.csv","a")
                         f.write(filename)
                         f.write("\n")
            with open('features3.csv') as csvfile:
                readCSV = csv.reader(csvfile, delimiter=' ')
                for row in readCSV:
                    audio_path = row[0]
                    y, sr = librosa.load(audio_path, sr=16000)
                    if get_data:
                        self.data.append((y, sr))
                        self.targets.append(classes[emotion])

                    with open('Audio3.csv') as csvfile:
                        readCSV = csv.reader(csvfile, delimiter=' ')
                        for row in readCSV:
                            audio1 = row[0]
                    if audio1[:6] in speak_test:
                        test.append(i)
                    else:
                        train.append(i)
                    i = i + 1
                self.train_sets.append(train)
                self.test_sets.append(test)
                get_data = False

[WakeUpYoimiya]

"if audio1[:6] in speak_test: test.append(i) ....... train.append(i)"
I think you can print out test[] and train[] here to check the data at first.

[AdamarisAinsley]

Name of the Wav file is Ses02F_impro01_F000.wav ,and after print train and test data here in the if else statement it is just printing [image: image.png] same as in the case of berlin [image: image.png] let me know things are working same or not?

…

On Mon, Sep 9, 2019 at 6:38 AM Terebi1996 ***@***.***> wrote: "if audio1[:6] in speak_test: test.append(i) ....... train.append(i)" I think you can print out test[] and train[] here to check the data at first. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#17?email_source=notifications&email_token=AMQHN5I3IIXJU4GKQBVWVI3QIWSIRA5CNFSM4H44NWU2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD6GAMDI#issuecomment-529270285>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AMQHN5MRJV3HF6IUGPHBEB3QIWSIRANCNFSM4H44NWUQ> .

[WakeUpYoimiya]

Your image file seems can't be shown here :(

[AdamarisAinsley]

IEMOCAP:
Berlin:

[WakeUpYoimiya]

you can add my discord. if u like

[AdamarisAinsley]

you can add my discord. if u like

what is your discord tag?