def predict():
# initialize the data dictionary that will be returned from the
# view
result = {"success": False}
# ensure an image was properly uploaded to our endpoint
if flask.request.method == "POST":
# initialize the data dictionary that will be returned from the
# view
# load wave file
f = flask.request.files["audio"]
#wav, _ = sf.read(io.BytesIO(file))
filename = datetime.now().strftime("%Y%m%d-%H%M%S") + ".wav"
# file = "./audioSamples/salli.wav"
f.save(secure_filename(filename))
wav, _ = librosa.load(filename, mono=True, sr=16000)
# get mfcc feature
mfcc = np.transpose(
np.expand_dims(librosa.feature.mfcc(wav, 16000), axis=0),
[0, 2, 1])
# run session
label = sess.run(y, feed_dict={x: mfcc})
result["predictions"] = []
# print label
r = data.print_index(label)
for index_list in label:
preds = data.index2str(index_list)
result["predictions"].append(preds)
# indicate that the request was a success
result["success"] = True
os.remove(filename)
# return the data dictionary as a JSON response
return flask.jsonify(result)
# CTC loss
loss = logit.sg_ctc(target=y, seq_len=seq_len)
decoded_sequence, _ = tf.nn.ctc_beam_search_decoder(
logit.sg_transpose(perm=[1, 0, 2]), seq_len, merge_repeated=False)
y = tf.sparse_to_dense(decoded_sequence[0].indices,
decoded_sequence[0].dense_shape,
decoded_sequence[0].values) + 1
#
# train
#0.0001
with tf.Session(config=tf.ConfigProto(allow_soft_placement=True)) as sess:
# init variables
tf.sg_init(sess)
# restore parameters
saver = tf.train.Saver()
saver.restore(sess, tf.train.latest_checkpoint('asset/train'))
#epoch[41182]-step[205919]
#tf.sg_train(lr=0, loss=get_loss(input=inputs, target=labels, seq_len=seq_len),
# ep_size=data.num_batch, max_ep=41182+5, sess=sess, max_keep=0, keep_interval=0, save_interval=0)
with tf.sg_queue_context():
for _ in xrange(5):
out = sess.run(y)
print_index(out)
decoded[0].values) + 1
# regcognize audio file
# perintah untuk menginput path file audio
tf.sg_arg_def(file=('', 'speech wave file to recognize.'))
# load audio file
file = sys.argv[1]
wav, sr = librosa.load(file, mono=True, sr=16000)
# mendapatkan mfcc feature
mfcc = np.transpose(np.expand_dims(librosa.feature.mfcc(wav, 16000), axis=0),
[0, 2, 1])
# run network
with tf.Session() as sess:
# init variables
tf.sg_init(sess)
# restore parameters
saver = tf.train.Saver()
saver.restore(sess, tf.train.latest_checkpoint('asset/train'))
# run session
label = sess.run(y, feed_dict={x: mfcc})
# print label
data.print_index(label)
# restore parameters
saver = tf.train.Saver(vars_to_train)
saver.restore(sess, tf.train.latest_checkpoint('asset/train'))
# run session
for i in xrange(10000):
new_loss, _, noise_out = sess.run([loss, optimizer, noise],
feed_dict={
x: mfccs[index],
targ: new_target.reshape((1, -1))
})
if i % 10 == 0:
print "iteration ", i #targ:corpus.daniter_label[index]
print new_loss
if i % 100 == 0:
label = sess.run(pred, feed_dict={x: mfccs[index]})
print index2str(label[0])
if index2str(label[0]) == fool:
break
label = sess.run(pred, feed_dict={x: mfccs[index]})
# print label
print_index(label)
print noise_out
np.save(fool + ".npy", noise_out + mfccs[0])
#TODO: find easier examples