12

我试图在 contrib 包(tf.contrib.ctc.ctc_loss)下使用 Tensorflow 的 CTC 实现,但没有成功。

  • 首先,有人知道我在哪里可以阅读一个好的分步教程吗?Tensorflow 的文档在这个主题上很差。
  • 我是否必须向 ctc_loss 提供带有交错空白标签的标签?
  • 即使使用长度为 1 的训练数据集超过 200 个 epoch,我也无法过度拟合我的网络。:(
  • 如何使用 tf.edit_distance 计算标签错误率?

这是我的代码:

with graph.as_default():

  max_length = X_train.shape[1]
  frame_size = X_train.shape[2]
  max_target_length = y_train.shape[1]

  # Batch size x time steps x data width
  data = tf.placeholder(tf.float32, [None, max_length, frame_size])
  data_length = tf.placeholder(tf.int32, [None])

  #  Batch size x max_target_length
  target_dense = tf.placeholder(tf.int32, [None, max_target_length])
  target_length = tf.placeholder(tf.int32, [None])

  #  Generating sparse tensor representation of target
  target = ctc_label_dense_to_sparse(target_dense, target_length)

  # Applying LSTM, returning output for each timestep (y_rnn1, 
  # [batch_size, max_time, cell.output_size]) and the final state of shape
  # [batch_size, cell.state_size]
  y_rnn1, h_rnn1 = tf.nn.dynamic_rnn(
    tf.nn.rnn_cell.LSTMCell(num_hidden, state_is_tuple=True, num_proj=num_classes), #  num_proj=num_classes
    data,
    dtype=tf.float32,
    sequence_length=data_length,
  )

  #  For sequence labelling, we want a prediction for each timestamp. 
  #  However, we share the weights for the softmax layer across all timesteps. 
  #  How do we do that? By flattening the first two dimensions of the output tensor. 
  #  This way time steps look the same as examples in the batch to the weight matrix. 
  #  Afterwards, we reshape back to the desired shape


  # Reshaping
  logits = tf.transpose(y_rnn1, perm=(1, 0, 2))

  #  Get the loss by calculating ctc_loss
  #  Also calculates
  #  the gradient.  This class performs the softmax operation for you, so    inputs
  #  should be e.g. linear projections of outputs by an LSTM.
  loss = tf.reduce_mean(tf.contrib.ctc.ctc_loss(logits, target, data_length))

  #  Define our optimizer with learning rate
  optimizer = tf.train.RMSPropOptimizer(learning_rate).minimize(loss)

  #  Decoding using beam search
  decoded, log_probabilities = tf.contrib.ctc.ctc_beam_search_decoder(logits, data_length, beam_width=10, top_paths=1)

谢谢!

更新(2016 年 6 月 29 日)

谢谢你,@jihyeon-seo!所以,我们在 RNN 的输入上有类似 [num_batch, max_time_step, num_features] 的东西。我们使用 dynamic_rnn 执行给定输入的循环计算,输出一个形状为 [num_batch, max_time_step, num_hidden] 的张量。之后,我们需要在每个 tilmestep 中使用权重共享进行仿射投影,因此我们必须重塑为 [num_batch*max_time_step, num_hidden],乘以形状为 [num_hidden, num_classes] 的权重矩阵,求和偏置撤消reshape, transpose(所以我们将有 [max_time_steps, num_batch, num_classes] 用于 ctc loss 输入),这个结果将作为 ctc_loss 函数的输入。我做的一切正确吗?

这是代码:

    cell = tf.nn.rnn_cell.MultiRNNCell([cell] * num_layers, state_is_tuple=True)

    h_rnn1, self.last_state = tf.nn.dynamic_rnn(cell, self.input_data, self.sequence_length, dtype=tf.float32)

    #  Reshaping to share weights accross timesteps
    x_fc1 = tf.reshape(h_rnn1, [-1, num_hidden])

    self._logits = tf.matmul(x_fc1, self._W_fc1) + self._b_fc1

    #  Reshaping
    self._logits = tf.reshape(self._logits, [max_length, -1, num_classes])

    #  Calculating loss
    loss = tf.contrib.ctc.ctc_loss(self._logits, self._targets, self.sequence_length)

    self.cost = tf.reduce_mean(loss)

更新(2016 年 7 月 11 日)

谢谢@Xiv。这是修复错误后的代码:

    cell = tf.nn.rnn_cell.MultiRNNCell([cell] * num_layers, state_is_tuple=True)

    h_rnn1, self.last_state = tf.nn.dynamic_rnn(cell, self.input_data, self.sequence_length, dtype=tf.float32)

    #  Reshaping to share weights accross timesteps
    x_fc1 = tf.reshape(h_rnn1, [-1, num_hidden])

    self._logits = tf.matmul(x_fc1, self._W_fc1) + self._b_fc1

    #  Reshaping
    self._logits = tf.reshape(self._logits, [-1, max_length, num_classes])
    self._logits = tf.transpose(self._logits, (1,0,2))

    #  Calculating loss
    loss = tf.contrib.ctc.ctc_loss(self._logits, self._targets, self.sequence_length)

    self.cost = tf.reduce_mean(loss)

更新 (07/25/16)

我在 GitHub 上发布了我的部分代码,只使用了一个语句。随意使用!:)

4

2 回答 2

6

我正在尝试做同样的事情。以下是我发现您可能感兴趣的内容。

很难找到 CTC 的教程,但是这个例子很有帮助

而对于空白标签,CTC 层假设空白索引为num_classes - 1,因此您需要为空白标签提供额外的类。

此外,CTC 网络执行 softmax 层。在您的代码中,RNN 层连接到 CTC 损失层。RNN层的输出是内部激活的,所以你需要再添加一个没有激活功能的隐藏层(可能是输出层),然后添加CTC损失层。

于 2016-06-28T07:04:10.287 回答
6

有关双向 LSTM、CTC 和编辑距离实现的示例,请参见此处,在 TIMIT 语料库上训练音素识别模型。如果您在该语料库的训练集上进行训练,您应该能够在 120 个 epoch 左右后将音素错误率降低到 20-25%。

于 2016-07-13T13:11:06.283 回答