model_filename = MODEL_FOLDER + model_id[k] + '.xml'
name, states, num_states, num_components, dim_observation, log_trans, log_coef, mean, log_var = ReadModel(
model_filename)
model = [log_trans, log_coef, mean, log_var]
models.append(model)
count = 0
sum = 0
# For each model
for k in range(len(model_id)):
# Load MFCC data
# To compute the global mean and log_var
MFCC_filename_list = glob(MFCC_FOLDER + '*' + model_id[k] + '*.txt')
feat_list = []
for filename in MFCC_filename_list:
feat_list.append(load(filename))
dim_observation = len(feat_list[0][0])
print('VDecode: ' + str(k + 1) + '/' + str(len(model_id)))
for i in range(len(feat_list)):
#print('VDecode: Using ' + str(i + 1) + '/' + str(len(feat_list)) + ' recording')
ans = VDecode(models, feat_list[i])
print('VDecode: predict = ' + str(ans) + ', real = ' + str(k))
# Wrong prediction
if ans != k:
count += 1
sum += 1
print('Error Rate = ' + str(float(count) / sum))
MAIN_DIR = getcwd()+'/'
WAVE_FOLDER = MAIN_DIR + 'wav/'
TEST_FOLDER = WAVE_FOLDER + 'test/'
CONFIG_DIR = MAIN_DIR + 'config/'
DICT_DIR = MAIN_DIR + 'dict/'
MFCC_DIR = MAIN_DIR + 'mfcc/test/'
MODEL_FOLDER = MAIN_DIR + 'model/'
if len(sys.argv) != 2:
sys.exit("Usage: Recognizer.py <dict>")
words, model_id = GetDictionary(DICT_DIR + sys.argv[1] + '.txt')
models = LoadModels(model_id,MODEL_FOLDER)
instruction = 'Get ready to speak (0~9) and press <Enter> to start record.\n Remember to leave 3 seconds of blank before and after the utterance.'
filename = TEST_FOLDER +'temp.wav'
Collect(filename,instruction)
signal = preprocess(filename)
cooked_filename = TEST_FOLDER+'cooked.wav'
signal.write(cooked_filename)
cook(cooked_filename,MFCC_DIR)
mfcc_filename = MFCC_DIR+split(cooked_filename)[1].replace('wav','txt')
mfcc = load(mfcc_filename)
ans = BWDecode(models,mfcc)
print('BWDecode: predict = '+words[ans])