Deploy

Mandarin audio.zip Above is my result.（the wave means “Thanks for BiaoBei, Thanks for the author’s work , thanks for community” in Mandarin) Now i want to deploy my model, so i saved model in pb format, but when i restoring pb model, I have encountered many problems. I stuck in these problem for two days.

here is restore code: if set mel and linear None, enqueue would throw error:

Traceback (most recent call last):
  File "TTSPythonWrap2.py", line 81, in <module>
    sess.run(enqueue, feed_dict=feed_dict)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 900, in run
    run_metadata_ptr)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1111, in _run
    str(subfeed_t.get_shape())))
ValueError: Cannot feed value of shape () for Tensor 'linear_targets:0', which has shape '(?, ?, 1025)'

if i set mel and linear an zero ndarray with shape [1,1,80], [1,1,1025], it seems “is_training” will be set True and throw error :

Traceback (most recent call last):
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1322, in _do_call
    return fn(*args)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1307, in _run_fn
    options, feed_dict, fetch_list, target_list, run_metadata)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1409, in _call_tf_sessionrun
    run_metadata)
tensorflow.python.framework.errors_impl.InvalidArgumentError: slice index 0 of dimension 1 out of bounds.
         [[Node: prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice = StridedSlice[Index=DT_INT32, T=DT_FLOAT, begin_mask=5, ellipsis_mask=0, end_mask=5, new_axis_mask=0, shrink_axis_mask=2, _device="/job:localhost/replica:0/task:0/device:GPU:0"](prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/Enter, prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/stack, prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/stack_1, prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/stack_2)]]
         [[Node: prefix/model/inference/encoder_cbhg/bidirectional_rnn/bw/bw/stack/_209 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_1776_prefix/model/inference/encoder_cbhg/bidirectional_rnn/bw/bw/stack", tensor_type=DT_INT32, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "TTSPythonWrap2.py", line 86, in <module>
    wav = sess.run(output)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 900, in run
    run_metadata_ptr)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1135, in _run
    feed_dict_tensor, options, run_metadata)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1316, in _do_run
    run_metadata)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1335, in _do_call
    raise type(e)(node_def, op, message)
tensorflow.python.framework.errors_impl.InvalidArgumentError: slice index 0 of dimension 1 out of bounds.
         [[Node: prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice = StridedSlice[Index=DT_INT32, T=DT_FLOAT, begin_mask=5, ellipsis_mask=0, end_mask=5, new_axis_mask=0, shrink_axis_mask=2, _device="/job:localhost/replica:0/task:0/device:GPU:0"](prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/Enter, prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/stack, prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/stack_1, prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/stack_2)]]
         [[Node: prefix/model/inference/encoder_cbhg/bidirectional_rnn/bw/bw/stack/_209 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_1776_prefix/model/inference/encoder_cbhg/bidirectional_rnn/bw/bw/stack", tensor_type=DT_INT32, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

Caused by op 'prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice', defined at:
  File "TTSPythonWrap2.py", line 44, in <module>
    name='prefix'
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/util/deprecation.py", line 432, in new_func
    return func(*args, **kwargs)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/importer.py", line 442, in import_graph_def
    _ProcessNewOps(graph)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/importer.py", line 234, in _ProcessNewOps
    for new_op in graph._add_new_tf_operations(compute_devices=False):  # pylint: disable=protected-access
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3563, in _add_new_tf_operations
    for c_op in c_api_util.new_tf_operations(self)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3563, in <listcomp>
    for c_op in c_api_util.new_tf_operations(self)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 3450, in _create_op_from_tf_operation
    ret = Operation(c_op, self)
  File "/root/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/ops.py", line 1740, in __init__
    self._traceback = self._graph._extract_stack()  # pylint: disable=protected-access

InvalidArgumentError (see above for traceback): slice index 0 of dimension 1 out of bounds.
         [[Node: prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice = StridedSlice[Index=DT_INT32, T=DT_FLOAT, begin_mask=5, ellipsis_mask=0, end_mask=5, new_axis_mask=0, shrink_axis_mask=2, _device="/job:localhost/replica:0/task:0/device:GPU:0"](prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/Enter, prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/stack, prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/stack_1, prefix/model/inference/decoder/while/BasicDecoderStep/TacoTrainingHelper/strided_slice/stack_2)]]
         [[Node: prefix/model/inference/encoder_cbhg/bidirectional_rnn/bw/bw/stack/_209 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_1776_prefix/model/inference/encoder_cbhg/bidirectional_rnn/bw/bw/stack", tensor_type=DT_INT32, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

restoring code:

from load import load_graph, load_graph_map
import tensorflow as tf
import numpy as np
from text import text_to_sequence
from util import audio


sentences = [
    "sil x ia4 v3 t ian1 l iou2 k e4 sp1 t ian1 l iou2 uo3 b u4 l iou2 sil"
]

fix_model_pb = '/home/sxdz/workspace/PycharmProj/tacotron-phone/frozen_model/tacotron-1.pb'
in_out_ops = ['datafeeder/inputs:0', 'datafeeder/input_lengths:0', 'model/stats/linear_outputs:0']


def debug_shape(t):
    print(t.get_shape().as_list())


with tf.Session() as sess:
    # tf.global_variables_initializer().run()
    # tf.local_variables_initializer().run()
    x = tf.placeholder(tf.int32, [None, None])
    x_len = tf.placeholder(tf.int32, [None])
    mel_t = tf.placeholder(tf.float32, [None, None, 80], 'mel_targets'),
    linear_t = tf.placeholder(tf.float32, [None, None, 1025], 'linear_targets')

    input_map = {
        'datafeeder/inputs:0': x,
        'datafeeder/input_lengths:0': x_len,
        'datafeeder/mel_targets:0': mel_t,
        'datafeeder/linear_targets:0': linear_t
    }
    with tf.gfile.FastGFile(fix_model_pb, 'rb') as f:
        graph_def = tf.GraphDef()

    graph_def.ParseFromString(f.read())

    sess.graph.as_default()

    tf.import_graph_def(
        graph_def,
        input_map=input_map,
        name='prefix'
    )


    # x = tf.get_default_graph().get_tensor_by_name('datafeeder/inputs:0')
    # x_len = tf.get_default_graph().get_tensor_by_name('datafeeder/input_lengths:0')
    # mel_t = tf.get_default_graph().get_tensor_by_name('datafeeder/mel_targets:0')
    # linear_tt = tf.get_default_graph().get_tensor_by_name('datafeeder/linear_targets:0')

    queue = tf.get_default_graph().get_operation_by_name('prefix/datafeeder/input_queue')
    enqueue = tf.get_default_graph().get_operation_by_name('prefix/datafeeder/input_queue_enqueue')
    dequeue = tf.get_default_graph().get_operation_by_name('prefix/datafeeder/input_queue_Dequeue')
    output = tf.get_default_graph().get_tensor_by_name('prefix/model/inference/dense/BiasAdd:0')

    cleaner_names = [x.strip() for x in ['transliteration_cleaners']]
    seq = text_to_sequence(sentences[0], cleaner_names)
    # # print(seq)
    x_mel = np.zeros((80,), dtype=np.float32).reshape((1, 1, 80))
    x_linear = np.zeros((1025,), dtype=np.float32).reshape((1, 1, 1025))
    feed_dict = {
        x: [np.asarray(seq, dtype=np.int32)],
        x_len: np.asarray([len(seq)], dtype=np.int32),
        **linear_t: x_linear,
        mel_t: x_mel,**
    }

    sess.run(queue)

    sess.run(enqueue, feed_dict=feed_dict)


    # sess.run(dequeue)
    wav = sess.run(output)

[sujeendran]

You can dump the graph def after loading it in synthesizer.py class by adding a extra line at the end of load function: tf.train.write_graph(self.session.graph.as_graph_def(), "models/", "graph.pb", as_text=True)

Then execute the tensorflow’s freeze_graph program to freeze it: $python freeze_graph.py --input_graph models/graph.pb --input_checkpoint logs-tacotron/model.ckpt-21000 --output_graph logs-tacotron/freezed.pb --output_node_names model/griffinlim/Squeeze

[sujeendran]

Hi, I made a class similar to synthesizer class with just few changes to run on the Deployment ready graph. This should work for you. This model has FIFO QUEUE for training only.

def load_graph(frozen_graph_filename):
    with tf.gfile.GFile(frozen_graph_filename, "rb") as f:
        graph_def = tf.GraphDef()
        graph_def.ParseFromString(f.read())
    with tf.Graph().as_default() as graph:
        # The name var will prefix every node name in graph if needed
        tf.import_graph_def(graph_def, name="")
    return graph

class Inferencer:  
  def load(self, model_path, model_name='tacotron'):
    print('Loading frozen graph for inference: %s' % model_path)
    graph = load_graph(model_path)
    self.input = graph.get_tensor_by_name('inputs:0')
    self.input_lengths = graph.get_tensor_by_name('input_lengths:0')
    self.wav_output = graph.get_tensor_by_name('model/griffinlim/Squeeze:0')
    #OR IF YOU ARE ATTACHING GRIFFINLIM AT INFERENCE TIME USE BELOW
    #self.wav_output = audio.inv_spectrogram_tensorflow(graph.get_tensor_by_name('model/inference/dense/BiasAdd:0')[0])
    self.session = tf.Session(graph = graph)