def get_action_param_pb(self, action_name, param_name, value):
"""
Converts an action parameter value to something readable by the switch.
The user provides a single value that will be passed to the action
function when the associated table entry is matched.
"""
p4info_param = self.get_action_param(action_name, param_name)
p4runtime_param = p4runtime_pb2.Action.Param()
p4runtime_param.param_id = p4info_param.id
p4runtime_param.value = encode(value, p4info_param.bitwidth)
return p4runtime_param
def build_packet_in(self, payload, metadata=None):
packet_in = p4runtime_pb2.PacketIn()
packet_in.payload = payload
if not metadata:
return packet_in
for name, value in metadata.items():
p4info_meta = self.get_packet_metadata("packet_in", name)
meta = packet_in.metadata.add()
meta.metadata_id = p4info_meta.id
meta.value = encode(value, p4info_meta.bitwidth)
return packet_in
def save_training_data(images, labels, filename):
assert images.shape[0] == labels.shape[0]
num_examples = images.shape[0]
with tf.python_io.TFRecordWriter(filename) as writer:
for index in range(num_examples):
image = images[index]
label = labels[index]
example = encode(image, label)
writer.write(example.SerializeToString())
def get_match_field_pb(self, table_name, match_field_name, value):
p4info_match = self.get_match_field(table_name, match_field_name)
bitwidth = p4info_match.bitwidth
p4runtime_match = p4runtime_pb2.FieldMatch()
p4runtime_match.field_id = p4info_match.id
match_type = p4info_match.match_type
#print(dir(p4info_pb2.MatchField))
if match_type == p4info_pb2.MatchField.VALID:
valid = p4runtime_match.valid
valid.value = bool(value)
elif match_type == p4info_pb2.MatchField.EXACT:
exact = p4runtime_match.exact
exact.value = encode(value, bitwidth)
elif match_type == p4info_pb2.MatchField.LPM:
lpm = p4runtime_match.lpm
lpm.value = encode(value[0], bitwidth)
lpm.prefix_len = value[1]
elif match_type == p4info_pb2.MatchField.TERNARY:
lpm = p4runtime_match.ternary
lpm.value = encode(value[0], bitwidth)
lpm.mask = encode(value[1], bitwidth)
elif match_type == p4info_pb2.MatchField.RANGE:
lpm = p4runtime_match.range
lpm.low = encode(value[0], bitwidth)
lpm.high = encode(value[1], bitwidth)
else:
raise Exception("Unsupported match type with type %r" % match_type)
return p4runtime_match
def get_match_field_pb(self, table_name, match_field_name, value):
"""
Converts a match field value to something readable by the switch.
For each match type, the user provides:
EXACT:
<value>
LPM:
[<string (ip address)>, <prefix length>]
TERNARY:
[<value>, <bit mask>]
RANGE:
[<low value>, <high value>]
"""
p4info_match = self.get_match_field(table_name, match_field_name)
bitwidth = p4info_match.bitwidth
p4runtime_match = p4runtime_pb2.FieldMatch()
p4runtime_match.field_id = p4info_match.id
match_type = p4info_match.match_type
if match_type == p4info_pb2.MatchField.EXACT:
exact = p4runtime_match.exact
exact.value = encode(value, bitwidth)
elif match_type == p4info_pb2.MatchField.LPM:
lpm = p4runtime_match.lpm
lpm.value = encode(value[0], bitwidth)
lpm.prefix_len = value[1]
elif match_type == p4info_pb2.MatchField.TERNARY:
lpm = p4runtime_match.ternary
lpm.value = encode(value[0], bitwidth)
lpm.mask = encode(value[1], bitwidth)
elif match_type == p4info_pb2.MatchField.RANGE:
lpm = p4runtime_match.range
lpm.low = encode(value[0], bitwidth)
lpm.high = encode(value[1], bitwidth)
else:
raise Exception("Unsupported match type with type %r" % match_type)
return p4runtime_match
def get_action_param_pb(self, action_name, param_name, value):
p4info_param = self.get_action_param(action_name, param_name)
p4runtime_param = p4runtime_pb2.Action.Param()
p4runtime_param.param_id = p4info_param.id
p4runtime_param.value = encode(value, p4info_param.bitwidth)
return p4runtime_param
def get_packetout_metadata_pb(self, metadata_name, value):
p4info_meta = self.get_packetout_meta("packet_out", metadata_name)
p4runtime_metadata = p4runtime_pb2.PacketMetadata()
p4runtime_metadata.metadata_id = p4info_meta.id
p4runtime_metadata.value = encode(value, p4info_meta.bitwidth)
return p4runtime_metadata
def main():
#---initialize---#
args = get_test_args()
HPS = Hps(args.hps_path)
hps = HPS.get_tuple()
trainer = get_trainer(args.hps_path, args.encoder_path, hps.g_mode,
hps.enc_mode)
if args.eval_t == 'None':
print(
'[Tacotron] - None is not a valid evaluation target! Please specify target manually, must be either V001, or V002.'
)
return
# Tacotron implementation: https://github.com/andi611/TTS-Tacotron-Pytorch
model = Tacotron(n_vocab=len(symbols),
embedding_dim=config.embedding_dim,
mel_dim=config.num_mels,
linear_dim=config.num_freq,
r=config.outputs_per_step,
padding_idx=config.padding_idx,
attention=config.attention,
use_mask=config.use_mask)
#---handle path---#
result_dir = os.path.join(args.result_dir, args.sub_result_dir)
os.makedirs(result_dir, exist_ok=True)
checkpoint_path = os.path.join(args.ckpt_dir, args.model_name)
if args.dataset == 'english' and not os.path.isdir(
'./ckpt_tacotron_english'):
print(
'[Tacotron] - Recommand using the following name for ckpt_dir: ./ckpt_tacotron_english/'
)
elif args.dataset == 'surprise' and not os.path.isdir(
'./ckpt_tacotron_surprise'):
print(
'[Tacotron] - Recommand using the following name for ckpt_dir: ./ckpt_tacotron_surprise/'
)
#---load and set model---#
print('[Tacotron] - Testing on the {} set.'.format(args.dataset))
print('[Tacotron] - Loading model: ', checkpoint_path)
checkpoint = torch.load(checkpoint_path)
model.load_state_dict(checkpoint["state_dict"])
#---load and set mappings---#
print('[Tacotron] - Loading mapping files: ', args.speaker2id_path)
valid_arguments(valid_target=args.dataset, arg=args.speaker2id_path)
with open(args.speaker2id_path, 'r') as f_json:
speaker2id = json.load(f_json)
print('[Tacotron] - Loading mapping files: ', args.multi2idx_path)
with open(args.multi2idx_path, 'r') as f_json:
multi2idx = json.load(f_json)
if not args.test_single:
#---parse testing list---#
print('[Tacotron] - Testing from list: ', args.synthesis_list)
valid_arguments(valid_target=args.dataset, arg=args.synthesis_list)
feeds = []
with open(args.synthesis_list, 'r') as f:
file = f.readlines()
for line in file:
line = line.split('\n')[0].split(' ')
feeds.append({
's_id': line[0].split('/')[1].split('_')[0],
'utt_id': line[0].split('/')[1].split('_')[1],
't_id': line[1],
})
print('[Tester] - Number of files to be resynthesize: ', len(feeds))
for feed in tqdm(feeds):
if feed['t_id'] == args.eval_t:
wav_path = os.path.join(
args.testing_dir,
feed['s_id'] + '_' + feed['utt_id'] + '.wav')
_, spec = get_spectrograms(wav_path)
encodings = encode(spec, trainer, hps.seg_len, save=False)
encodings = parse_encodings(encodings)
line = ''.join([multi2idx[encoding] for encoding in encodings])
print(line)
out_path = os.path.join(
result_dir, feed['t_id'] + '_' + feed['utt_id'] + '.wav')
synthesis_speech(model, text=line, path=out_path)
else:
wav_path = './data/english/train/voice/V002_0674932509.wav'
# wav_path = './data/english/train/voice/V002_2252538703.wav'
# wav_path = './data/english/train/voice/V002_1665800749.wav'
_, spec = get_spectrograms(wav_path)
encodings = encode(spec, trainer, hps.seg_len, save=False)
write_encodings(path='./result/result.wav', encodings=encodings)
parsed_encodings = parse_encodings(encodings)
line = ''.join([multi2idx[encoding] for encoding in parsed_encodings])
print(line)
synthesis_speech(model, text=line, path='./result/result.wav')
# model.decoder.max_decoder_steps = config.max_decoder_steps # Set large max_decoder steps to handle long sentence outputs
sys.exit(0)
