def load(self, checkpoint_path, model_name='tacotron'): print('Constructing model: %s' % model_name) inputs = tf.placeholder(tf.int32, [1, None], 'inputs') input_lengths = tf.placeholder(tf.int32, [1], 'input_lengths') identity = tf.placeholder(tf.int32, [1], 'identity') with tf.variable_scope('model') as scope: hparams.chinese_symbol = True hparams.max_iters = 400 self.model = create_model(model_name, hparams) reader2 = pywrap_tensorflow.NewCheckpointReader(checkpoint_path) var_to_shape_map = reader2.get_variable_to_shape_map() id_num = var_to_shape_map['model/inference/embedding_id'][0] self.model.initialize(inputs, input_lengths, identities=identity, id_num=id_num) self.wav_output = audio.inv_spectrogram_tensorflow( self.model.linear_outputs[0]) self.alignment = self.model.alignments[0] print('Loading checkpoint: %s' % checkpoint_path) self.session = tf.Session() self.session.run(tf.global_variables_initializer()) saver = tf.train.Saver() saver.restore(self.session, checkpoint_path)
def load(self, checkpoint_path, reference_mel=None, model_name='tacotron'): print('Constructing model: %s' % model_name) inputs = tf.placeholder(tf.int32, [1, None], 'inputs') input_lengths = tf.placeholder(tf.int32, [1], 'input_lengths') if reference_mel is not None: reference_mel = tf.placeholder(tf.float32, [1, None, hparams.num_mels], 'reference_mel') # Only used in teacher-forcing generating mode if self.teacher_forcing_generating: mel_targets = tf.placeholder(tf.float32, [1, None, hparams.num_mels], 'mel_targets') else: mel_targets = None with tf.variable_scope('model') as scope: self.model = create_model(model_name, hparams) self.model.initialize(inputs, input_lengths, mel_targets=mel_targets, reference_mel=reference_mel) self.wav_output = audio.inv_spectrogram_tensorflow( self.model.linear_outputs[0]) self.alignments = self.model.alignments[0] print('Loading checkpoint: %s' % checkpoint_path) self.session = tf.Session() self.session.run(tf.global_variables_initializer()) saver = tf.train.Saver() saver.restore(self.session, checkpoint_path)
def load(self, checkpoint_path, model_name='tacotron'): print('Constructing model: %s' % model_name) inputs = tf.placeholder(tf.int32, [1, None], 'inputs') input_lengths = tf.placeholder(tf.int32, [1], 'input_lengths') mel_spec = tf.placeholder(tf.float32, [None, None, self.hparams.num_mels], 'mel_targets') with tf.variable_scope('model') as scope: if self.hparams.enable_fv1 or self.hparams.enable_fv2: self.net = ResCNN(data=mel_spec, hyparam=self.hparams) self.net.inference() voice_print_feature = tf.reduce_mean(self.net.features, 0) else: voice_print_feature = None self.model = create_model(model_name, self.hparams) self.model.initialize(inputs=inputs, input_lengths=input_lengths, voice_print_feature=voice_print_feature) self.wav_output = audio.inv_spectrogram_tensorflow( self.model.linear_outputs[0]) print('Loading checkpoint: %s' % checkpoint_path) self.session = tf.Session() self.session.run(tf.global_variables_initializer()) saver = tf.train.Saver() saver.restore(self.session, checkpoint_path)
def update(self): with tf.variable_scope('model') as scope: self.model.update(hparams) self.wav_output = audio.inv_spectrogram_tensorflow( self.model.linear_outputs[0]) print('Loading checkpoint: %s' % self.checkpoint_path) self.session = tf.Session() self.session.run(tf.global_variables_initializer()) saver = tf.train.Saver() saver.restore(self.session, self.checkpoint_path)
def load(self, checkpoint_path, model_name='tacotron'): inputs = tf.placeholder(tf.int32, [1, None], 'inputs') input_lengths = tf.placeholder(tf.int32, [1], 'input_lengths') with tf.variable_scope('model') as scope: self.model = Tacotron(hparams) self.model.initialize(inputs, input_lengths) self.wav_output = audio.inv_spectrogram_tensorflow(self.model.linear_outputs[0]) # 读取已有模型 self.session = tf.Session() self.session.run(tf.global_variables_initializer()) saver = tf.train.Saver() saver.restore(self.session, checkpoint_path)
def load(self, checkpoint_path, model_name='tacotron'): print('Constructing model: %s' % model_name) inputs = tf.placeholder(tf.int32, [1, None], 'inputs') input_lengths = tf.placeholder(tf.int32, [1], 'input_lengths') with tf.variable_scope('model') as scope: self.model = create_model(model_name, hparams) self.model.initialize(inputs, input_lengths) self.wav_output = audio.inv_spectrogram_tensorflow(self.model.linear_outputs[0]) print('Loading checkpoint: %s' % checkpoint_path) self.session = tf.Session() self.session.run(tf.global_variables_initializer()) saver = tf.train.Saver() saver.restore(self.session, checkpoint_path)
def load(self, checkpoint_path, vgg19_path, model_name='tacotron'): print('Constructing model: %s' % model_name) inputs = tf.placeholder(tf.float32, [1, hparams.image_dim, hparams.image_dim, 3], 'inputs') with tf.variable_scope('model') as _: self.model = create_model(model_name, hparams) self.model.initialize(inputs, vgg19_path) self.wav_output = audio.inv_spectrogram_tensorflow( self.model.linear_outputs[0]) print('Loading checkpoint: %s' % checkpoint_path) self.session = tf.Session() self.session.run(tf.global_variables_initializer()) checkpoint_saver = tf.train.import_meta_graph( '%s.%s' % (checkpoint_path, 'meta')) checkpoint_saver.restore(self.session, checkpoint_path)
def load(self, checkpoint_path, model_name='tacotron'): print('Constructing model: %s' % model_name) inputs = tf.placeholder(tf.int32, [1, None], 'inputs') input_lengths = tf.placeholder(tf.int32, [1], 'input_lengths') with tf.variable_scope('model') as scope: self.model = create_model(model_name, hparams) self.model.initialize(inputs, input_lengths) self.wav_output = audio.inv_spectrogram_tensorflow( self.model.linear_outputs[0]) print('Loading checkpoint: %s' % checkpoint_path) config = tf.ConfigProto() # making core utilization optimal, low value more optimaztion config.gpu_options.per_process_gpu_memory_fraction = 0.2 self.session = tf.Session(config=config) self.session.run(tf.global_variables_initializer()) saver = tf.train.Saver() saver.restore(self.session, checkpoint_path)
import tensorflow as tf from models import create_model from util import audio from hparams import hparams checkpoint_path = "/tacotron-20180906/model.ckpt" model_name = 'tacotron' inputs = tf.placeholder(tf.int32, [1, None], 'inputs') input_lengths = tf.placeholder(tf.int32, [1], 'input_lengths') with tf.variable_scope('model') as scope: model = create_model(model_name, hparams) model.initialize(inputs, input_lengths) wav_output = audio.inv_spectrogram_tensorflow(model.linear_outputs[0]) print('Loading checkpoint: %s' % checkpoint_path) session = tf.Session() session.run(tf.global_variables_initializer()) saver = tf.train.Saver() saver.restore(session, checkpoint_path) with open("eval.pb", 'w') as f: g = tf.get_default_graph() f.write(str(g.as_graph_def()))