def evaluate(hyper_params):
if hyper_params["test_dataset_dirs"] is None:
logging.fatal("Setting test_dataset_dirs in config file is mandatory for evaluation mode")
return
# Load the test set data
data_processor = dataprocessor.DataProcessor(hyper_params["test_dataset_dirs"])
test_set = data_processor.get_dataset()
logging.info("Using %d size of test set", len(test_set))
if len(test_set) == 0:
logging.fatal("No files in test set during an evaluation mode")
return
with tf.Session() as sess:
# create model
model = AcousticModel(hyper_params["num_layers"], hyper_params["hidden_size"], hyper_params["batch_size"],
hyper_params["max_input_seq_length"], hyper_params["max_target_seq_length"],
hyper_params["input_dim"], hyper_params["batch_normalization"],
hyper_params["char_map_length"])
model.create_forward_rnn()
model.initialize(sess)
model.restore(sess, hyper_params["checkpoint_dir"] + "/acoustic/")
wer, cer = model.evaluate_full(sess, test_set, hyper_params["max_input_seq_length"],
hyper_params["signal_processing"], hyper_params["char_map"])
print("Resulting WER : {0:.3g} %".format(wer))
print("Resulting CER : {0:.3g} %".format(cer))
return
def record_and_write(audio_processor, hyper_params):
import pyaudio
_CHUNK = hyper_params["max_input_seq_length"]
_SR = 22050
p = pyaudio.PyAudio()
with tf.Session() as sess:
# create model
model = AcousticModel(hyper_params["num_layers"], hyper_params["hidden_size"], 1,
hyper_params["max_input_seq_length"], hyper_params["max_target_seq_length"],
hyper_params["input_dim"], hyper_params["batch_normalization"],
hyper_params["char_map_length"])
model.create_forward_rnn()
model.initialize(sess)
model.restore(sess, hyper_params["checkpoint_dir"] + "/acoustic/")
# Create stream of listening
stream = p.open(format=pyaudio.paInt16, channels=1, rate=_SR, input=True, frames_per_buffer=_CHUNK)
print("NOW RECORDING...")
while True:
data = stream.read(_CHUNK)
data = np.fromstring(data)
feat_vec, original_feat_vec_length = audio_processor.process_signal(data, _SR)
(a, b) = feat_vec.shape
feat_vec = feat_vec.reshape((a, 1, b))
predictions = model.process_input(sess, feat_vec, [original_feat_vec_length])
result = [dataprocessor.DataProcessor.get_labels_str(hyper_params["char_map"], prediction)
for prediction in predictions]
print(result, end="")
def record_and_write(audio_processor, hyper_params):
import pyaudio
_CHUNK = hyper_params["max_input_seq_length"]
_SR = 22050
p = pyaudio.PyAudio()
with tf.Session() as sess:
# create model
model = AcousticModel(hyper_params["num_layers"], hyper_params["hidden_size"], 1,
hyper_params["max_input_seq_length"], hyper_params["max_target_seq_length"],
hyper_params["input_dim"], hyper_params["batch_normalization"],
language=hyper_params["language"])
model.create_forward_rnn()
model.initialize(sess)
model.restore(sess, hyper_params["checkpoint_dir"])
# Create stream of listening
stream = p.open(format=pyaudio.paInt16, channels=1, rate=_SR, input=True, frames_per_buffer=_CHUNK)
print("NOW RECORDING...")
while True:
data = stream.read(_CHUNK)
data = np.fromstring(data)
feat_vec, original_feat_vec_length = audio_processor.process_signal(data, _SR)
(a, b) = feat_vec.shape
feat_vec = feat_vec.reshape((a, 1, b))
result = model.process_input(sess, feat_vec, [original_feat_vec_length])
print(result, end="")
def process_file(audio_processor, hyper_params, file):
feat_vec, original_feat_vec_length = audio_processor.process_audio_file(file)
if original_feat_vec_length > hyper_params["max_input_seq_length"]:
logging.warning("File too long")
return
elif original_feat_vec_length < hyper_params["max_input_seq_length"]:
# Pad the feat_vec with zeros
pad_length = hyper_params["max_input_seq_length"] - original_feat_vec_length
padding = np.zeros((pad_length, hyper_params["input_dim"]), dtype=np.float)
feat_vec = np.concatenate((feat_vec, padding), 0)
with tf.Session() as sess:
# create model
model = AcousticModel(hyper_params["num_layers"], hyper_params["hidden_size"], 1,
hyper_params["max_input_seq_length"], hyper_params["max_target_seq_length"],
hyper_params["input_dim"], hyper_params["batch_normalization"],
hyper_params["char_map_length"])
model.create_forward_rnn()
model.initialize(sess)
model.restore(sess, hyper_params["checkpoint_dir"] + "/acoustic/")
(a, b) = feat_vec.shape
feat_vec = feat_vec.reshape((a, 1, b))
predictions = model.process_input(sess, feat_vec, [original_feat_vec_length])
transcribed_text = [dataprocessor.DataProcessor.get_labels_str(hyper_params["char_map"], prediction)
for prediction in predictions]
print(transcribed_text[0])
def evaluate(hyper_params):
if hyper_params["test_dataset_dirs"] is None:
logging.fatal("Setting test_dataset_dirs in config file is mandatory for evaluation mode")
return
# Load the test set data
data_processor = dataprocessor.DataProcessor(hyper_params["test_dataset_dirs"])
test_set = data_processor.get_dataset()
logging.info("Using %d size of test set", len(test_set))
if len(test_set) == 0:
logging.fatal("No files in test set during an evaluation mode")
return
with tf.Session() as sess:
# create model
model = AcousticModel(hyper_params["num_layers"], hyper_params["hidden_size"], hyper_params["batch_size"],
hyper_params["max_input_seq_length"], hyper_params["max_target_seq_length"],
hyper_params["input_dim"], hyper_params["batch_normalization"],
language=hyper_params["language"])
model.create_forward_rnn()
model.initialize(sess)
model.restore(sess, hyper_params["checkpoint_dir"])
wer, cer = model.evaluate_full(sess, test_set, hyper_params["max_input_seq_length"],
hyper_params["signal_processing"])
print("Resulting WER : {0:.3g} %".format(wer))
print("Resulting CER : {0:.3g} %".format(cer))
return
def process_file(audio_processor, hyper_params, file):
feat_vec, original_feat_vec_length = audio_processor.process_audio_file(file)
if original_feat_vec_length > hyper_params["max_input_seq_length"]:
logging.warning("File too long")
return
elif original_feat_vec_length < hyper_params["max_input_seq_length"]:
# Pad the feat_vec with zeros
pad_length = hyper_params["max_input_seq_length"] - original_feat_vec_length
padding = np.zeros((pad_length, hyper_params["input_dim"]), dtype=np.float)
feat_vec = np.concatenate((feat_vec, padding), 0)
with tf.Session() as sess:
# create model
model = AcousticModel(hyper_params["num_layers"], hyper_params["hidden_size"], 1,
hyper_params["max_input_seq_length"], hyper_params["max_target_seq_length"],
hyper_params["input_dim"], hyper_params["batch_normalization"],
hyper_params["char_map_length"])
model.create_forward_rnn()
model.initialize(sess)
model.restore(sess, hyper_params["checkpoint_dir"] + "/acoustic/")
(a, b) = feat_vec.shape
feat_vec = feat_vec.reshape((a, 1, b))
predictions = model.process_input(sess, feat_vec, [original_feat_vec_length])
transcribed_text = [dataprocessor.DataProcessor.get_labels_str(hyper_params["char_map"], prediction)
for prediction in predictions]
print(transcribed_text[0])
def test_create_forward_rnn(self):
tf.reset_default_graph()
with tf.Session():
model = AcousticModel(self.num_layers, self.hidden_size,
self.batch_size, self.max_input_seq_length,
self.max_target_seq_length, self.input_dim,
self.normalization, self.num_labels)
model.create_forward_rnn()
def process_file(audio_processor, hyper_params, file):
feat_vec, original_feat_vec_length = audio_processor.process_audio_file(file)
if original_feat_vec_length > hyper_params["max_input_seq_length"]:
logging.warning("File too long")
return
with tf.Session() as sess:
# create model
model = AcousticModel(hyper_params["num_layers"], hyper_params["hidden_size"], 1,
hyper_params["max_input_seq_length"], hyper_params["max_target_seq_length"],
hyper_params["input_dim"], hyper_params["batch_normalization"],
language=hyper_params["language"])
model.create_forward_rnn()
model.initialize(sess)
model.restore(sess, hyper_params["checkpoint_dir"])
(a, b) = feat_vec.shape
feat_vec = feat_vec.reshape((a, 1, b))
transcribed_text = model.process_input(sess, feat_vec, [original_feat_vec_length])
print(transcribed_text[0])
def test_create_forward_rnn(self):
tf.reset_default_graph()
with tf.Session():
model = AcousticModel(self.num_layers, self.hidden_size, self.batch_size, self.max_input_seq_length,
self.max_target_seq_length, self.input_dim, self.normalization, self.num_labels)
model.create_forward_rnn()
