def _update_model_name(self):
"""
Generates a unique model name based on parameters passed and put it on `self.model_name`.
This is used when saving the model.
"""
# get first letters of emotions, for instance:
# ["sad", "neutral", "happy"] => 'HNS' (sorted alphabetically)
emotions_str = get_first_letters(self.emotions)
# 'c' for classification & 'r' for regression
problem_type = 'c' if self.classification else 'r'
dropout_str = get_dropout_str(self.dropout, n_layers=self.n_dense_layers + self.n_rnn_layers)
self.model_name = f"{emotions_str}-{problem_type}-{self.cell.__name__}-layers-{self.n_rnn_layers}-{self.n_dense_layers}-units-{self.rnn_units}-{self.dense_units}-dropout-{dropout_str}.h5"
def load_metadata_from_desc_file(self, desc_files, partition):
"""Read metadata from a CSV file & Extract and loads features of audio files
Params:
desc_files (list): list of description files (csv files) to read from
partition (str): whether is "train" or "test"
"""
# empty dataframe
df = pd.DataFrame({'path': [], 'emotion': []})
for desc_file in desc_files:
# concat dataframes
df = pd.concat((df, pd.read_csv(desc_file)), sort=False)
if self.verbose:
print("[*] Loading audio file paths and its corresponding labels...")
# get columns
audio_paths, emotions = list(df['path']), list(df['emotion'])
# if not classification, convert emotions to numbers
if not self.classification:
# so naive and need to be implemented
# in a better way
if len(self.emotions) == 3:
self.categories = {'sad': 1, 'neutral': 2, 'happy': 3}
elif len(self.emotions) == 5:
self.categories = {'angry': 1, 'sad': 2, 'neutral': 3, 'ps': 4, 'happy': 5}
else:
raise TypeError("Regression is only for either ['sad', 'neutral', 'happy'] or ['angry', 'sad', 'neutral', 'ps', 'happy']")
emotions = [ self.categories[e] for e in emotions ]
# make features folder if does not exist
if not os.path.isdir(self.features_folder_name):
os.mkdir(self.features_folder_name)
# get label for features
label = get_label(self.audio_config)
# construct features file name
n_samples = len(audio_paths)
first_letters = get_first_letters(self.emotions)
name = os.path.join(self.features_folder_name, f"{partition}_{label}_{first_letters}_{n_samples}.npy")
if os.path.isfile(name):
# if file already exists, just load then
if self.verbose:
print("[+] Feature file already exists, loading...")
features = np.load(name)
else:
# file does not exist, extract those features and dump them into the file
features = []
append = features.append
for audio_file in tqdm.tqdm(audio_paths, f"Extracting features for {partition}"):
feature = extract_feature(audio_file, **self.audio_config)
if self.input_dimension is None:
self.input_dimension = feature.shape[0]
append(feature)
# convert to numpy array
features = np.array(features)
# save it
np.save(name, features)
if partition == "train":
try:
self.train_audio_paths
except AttributeError:
self.train_audio_paths = audio_paths
self.train_emotions = emotions
self.train_features = features
else:
if self.verbose:
print("[*] Adding additional training samples")
self.train_audio_paths += audio_paths
self.train_emotions += emotions
self.train_features = np.vstack((self.train_features, features))
elif partition == "test":
try:
self.test_audio_paths
except AttributeError:
self.test_audio_paths = audio_paths
self.test_emotions = emotions
self.test_features = features
else:
if self.verbose:
print("[*] Adding additional testing samples")
self.test_audio_paths += audio_paths
self.test_emotions += emotions
self.test_features = np.vstack((self.test_features, features))
else:
raise TypeError("Invalid partition, must be either train/test")
def _update_model_name(self):
emotions_str = get_first_letters(self.emotions)
# 'c' for classification & 'r' for regression
problem_type = 'c' if self.classification else 'r'
dropout_str = get_dropout_str(self.dropout, n_layers=self.n_dense_layers + self.n_rnn_layers)
self.model_name = f"{emotions_str}-{problem_type}-{self.cell.__name__}-layers-{self.n_rnn_layers}-{self.n_dense_layers}-units-{self.rnn_units}-{self.dense_units}-dropout-{dropout_str}.h5"