Cast hparam to the provided type, if compatible. Args: name: Name of the hparam to be cast. param_type: The type of the hparam.

feed-forward decoder Args: encoder_output: encoder outputs, shape: [batch, x_steps, d_model] duration_indexes: argmax

Helper function for raising a value error for bad assignment.

Helper function for raising a value error for reuse of name.

clip norm using tf.clip_by_norm

used for get logit length

Generate a square mask for the sequence. The masked positions are filled with bool(True). Unmasked positions are filled with bool(False).

NOTE: Changes to the returned TF_Tensor are not reflected in the actual device memory!

NOTE: Changes to the returned TF_Tensor are not reflected in the actual device memory!

Hertz to Bark conversion */

parse the jsonfile, output the parameters

reset num_err and num_total to zero

Discard end symbol and elements after end symbol Args: seqs: shape=(batch_size, seq_length) eos: eos id Returns: val

Returns the window-shift in samples, after resampling.

Forward just one chunk Args: xs (tf.Tensor), shape=[B(1), T, feat_dim, 1]: chunk input offset (int): current offset

Add a task

complex number absolute value square

convert a list of ids to a sentence

Hertz to Mel conversion */

restore from the best model

creates a Dataset that includes only 1/num_shards of this dataset

Split the last dimension into (num_heads, depth). Transpose the result such that the shape is (batch_size, num_heads, seq_len, depth)

Split the last dimension into (num_heads, depth). Transpose the result such that the shape is (batch_size, num_heads, seq_len, depth)

Accumulate errors and counts

return int in milliseconds

return gcd */