paddlespeech.vector.io.embedding_norm module
- class paddlespeech.vector.io.embedding_norm.InputNormalization(mean_norm=True, std_norm=True, norm_type='global')[source]
Bases:
object
Methods
__call__
(x, lengths[, spk_ids, stop_gradient])Returns the tensor with the surrounding context. Args: x (paddle.Tensor): A batch of tensors. lengths (paddle.Tensor): A batch of tensors containing the relative length of each sentence (e.g, [0.7, 0.9, 1.0]). It is used to avoid computing stats on zero-padded steps. spk_ids (paddle.Tensor, optional): tensor containing the ids of each speaker (e.g, [0 10 6]). It is used to perform per-speaker normalization when norm_type='speaker'. Defaults to paddle.to_tensor([], dtype="float32"). Returns: paddle.Tensor: The normalized feature or embedding.
save
(path)Save statistic dictionary.
to
(device)Puts the needed tensors in the right device.
- save(path)[source]
Save statistic dictionary.
- Args:
path (str): A path where to save the dictionary.
- spk_dict_count: Dict[int, int]
- spk_dict_mean: Dict[int, Tensor]
- spk_dict_std: Dict[int, Tensor]