print('True transcription:', text_true.lower())
#text_true.translate(str.maketrans('', '', string.punctuation)).lower()
from scipy.special import softmax
import numpy as np
index_map = {'neg': 0, 'neu': 1, 'pos': 2}
rev = { v:k for k,v in index_map.items()}
##### audio
# convert audio into feature
audio = audio2wave(file)
audio_X = prepare_data_librosa(audio,
features='logmel',
scaled=True)
stackedX_test = stacked_dataset(all_models, audio_X)
audio_pred = audio_logistic.predict(stackedX_test)
print('Audio prediction:', [rev[item] for item in audio_pred])
##### text
text_pred = predict_text(bert_model,text)
print('Text prediction:', [rev[item] for item in np.array([np.argmax(text_pred)])])
##### ensemble
ensemble_text_test = softmax(text_pred)
ensemble_audio_test = audio_logistic.predict_proba(stackedX_test)
stack_test = np.dstack((ensemble_text_test, ensemble_audio_test))
### audio data################################################################
import librosa
from tqdm import tqdm
ensemble_audio = []
for i in tqdm(ensemble_list):
y, sr = librosa.load(i, sr=16000, duration=7.5)
ensemble_audio.append(y)
from util.prepare_data import prepare_data_librosa
N_EMOTIONS = 3 # 3 or 4
FEATURE = 'logmel' #logmel or mfcc
scaled = True # True or False
audio_X = prepare_data_librosa(ensemble_audio,
features=FEATURE,
scaled=scaled)
def cate(array):
new_array = []
for i in array:
if i == 'hap':
i = 'pos'
elif i == 'neu':
i = 'neu'
else:
i = 'neg'
new_array.append(i)
return (new_array)
audio_hap = []
audio_hap.append(audios_db[0])
audio_neu = []
audio_neu.append(audios_db[1])
audio_ang = []
audio_ang.append(audios_db[2])
audio_sad = []
audio_sad.append(audios_db[5])
#%%
from util.prepare_data import prepare_data_librosa
FEATURE = 'logmel'
hap_1, hap_2, hap_3 = prepare_data_librosa(audio_hap,
features=FEATURE,
N_FEATURES=3,
scaled=True)
neu_1, neu_2, neu_3 = prepare_data_librosa(audio_neu,
features=FEATURE,
N_FEATURES=3,
scaled=True)
ang_1, ang_2, ang_3 = prepare_data_librosa(audio_ang,
features=FEATURE,
N_FEATURES=3,
scaled=True)
sad_1, sad_2, sad_3 = prepare_data_librosa(audio_sad,
features=FEATURE,
N_FEATURES=3,
scaled=True)