def country_model():
audio_wav = request.data
mfcc = audio_mfcc(audio_wav)
# Gender Model
gender_output, gender_prob = generate_pred(mel=mfcc, model=model_gender,
label=CommonVoiceModels.Gender.OUTPUT,
model_name=CommonVoiceModels.Gender,
)
return {'Prediction': gender_output, 'Probability': gender_prob}, 201
def convert_to_wav(self, index) -> None:
clips_name = self.label_data.path.values[index]
path = os.path.join(self.clips_dir, clips_name)
sample_length_in_seconds = 1
try:
signal, _ = librosa.load(path=path, sr=self.SAMPLE_RATE)
signal, _ = librosa.effects.trim(
signal, top_db=config.CommonVoiceModels.Frame.FRAME['TOP_DB'])
duration = len(signal) // self.SAMPLE_RATE
start = 0
step = int(sample_length_in_seconds * self.SAMPLE_RATE)
for i in range(1, duration + 1):
data = signal[start:start + step]
training_mfcc = audio_mfcc(data)
assert training_mfcc.shape[0] == self.model.PARAM["INPUT_SIZE"]
assert training_mfcc.shape[1] == 13
clip_name = "{}".format(clips_name.split(".")[0])
label_name = self.label_data[self.label_data.path ==
clips_name][
self.data_label].values[0]
if label_name in config.DO_NOT_INCLUDE:
break
train_test_choice = np.random.choice(
["train_data", "val_data", "test_data"], p=[0.7, 0.2, 0.1])
dir_path = os.path.join(self.document_path, self.data_label,
train_test_choice)
check_dir(dir_path)
dir_path = os.path.join(self.document_path, self.data_label,
train_test_choice, label_name)
check_dir(dir_path)
save_path = os.path.join(dir_path, clip_name + "_" + str(i))
np.save(save_path, training_mfcc.T)
start = self.SAMPLE_RATE * i
self.add_count += 1
except FileNotFoundError:
_logger.info("Can't find the file {}".format(clips_name))
self.remove_count += 1
except FileExistsError:
_logger.warning("Error in creating folder that's already created")
def load_audio_wav(audio_wav):
signal, _ = librosa.load(audio_wav, sr=SAMPLE_RATE)
mfcc = audio_mfcc(signal)
return mfcc