espnet2.tts.feats_extract.abs_feats_extract.AbsFeatsExtract

source

class espnet2.tts.feats_extract.abs_feats_extract.AbsFeatsExtract(*args, **kwargs)

Bases: Module, ABC

Abstract base class for feature extraction modules in the ESPnet TTS framework.

This class serves as a blueprint for creating feature extraction components. Subclasses must implement the methods defined here to provide specific functionality for extracting features from audio inputs.

None

output_size() → int

Returns the size of the output features.

get_parameters() → Dict[str, Any]

Returns the parameters of the feature extraction module.

forward(input

torch.Tensor, input_lengths: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]: Processes the input tensor and returns the extracted features along with the lengths of the input sequences.

• Raises:NotImplementedError – If a subclass does not implement any of the abstract methods.

########### Examples

To create a concrete implementation of this class, subclass it and implement the abstract methods as follows:

class MyFeatureExtractor(AbsFeatsExtract): : def output_size(self) -> int: : return 128 # Example output size <br/> def get_parameters(self) -> Dict[str, Any]: : return {“param1”: 1, “param2”: 2} # Example parameters <br/> def forward(self, input: torch.Tensor, input_lengths: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]: <br/>

Example feature extraction logic

features = input.mean(dim=1, keepdim=True) # Simplified example return features, input_lengths

Initialize internal Module state, shared by both nn.Module and ScriptModule.

abstract forward(input: Tensor, input_lengths: Tensor) → Tuple[Tensor, Tensor]

Computes the forward pass of the feature extraction module.

This method takes an input tensor and its corresponding lengths, and processes them to produce the extracted features along with their lengths.

• Parameters:
  • input (torch.Tensor) – A tensor containing the input data.
  • input_lengths (torch.Tensor) – A tensor containing the lengths of the input sequences.
• Returns: A tuple containing: : - A tensor of extracted features.
  • A tensor of lengths for the extracted features.
• Return type: Tuple[torch.Tensor, torch.Tensor]
• Raises:NotImplementedError – If the method is not implemented in a subclass.

########### Examples

>>> model = SomeConcreteFeatsExtractModel()
>>> input_tensor = torch.randn(10, 20)  # Example input
>>> lengths = torch.tensor([20] * 10)  # All sequences have length 20
>>> features, lengths = model.forward(input_tensor, lengths)
>>> print(features.shape)  # Expected output shape based on model design

abstract get_parameters() → Dict[str, Any]

Retrieves the parameters of the feature extraction module.

This method should be implemented by subclasses to return a dictionary containing the relevant parameters for the feature extraction process. The parameters may include configuration settings, hyperparameters, or any other relevant information needed to initialize or replicate the feature extraction behavior.

• Returns: A dictionary where the keys are parameter names (as strings) and the values are the corresponding parameter values (of any type).
• Raises:NotImplementedError – If the method is not implemented in a subclass.

########### Examples

If a subclass implements this method, it might return a dictionary like the following:

>>> extractor = MyFeatureExtractor()
>>> params = extractor.get_parameters()
>>> print(params)
{'learning_rate': 0.001, 'num_layers': 3}

abstract output_size() → int

Abstract base class for feature extraction modules.

This class defines the interface for feature extraction modules, including methods to obtain the output size, parameters, and perform the forward pass. Subclasses must implement the abstract methods defined in this class.

None

output_size()

Returns the size of the output features.

get_parameters()

Returns the parameters of the model as a dictionary.

forward()

Performs the forward pass of the model on the input tensor.

• Raises:NotImplementedError – If the method is not implemented in a subclass.

########### Examples

class MyFeatureExtractor(AbsFeatsExtract): : def output_size(self) -> int: : return 256 <br/> def get_parameters(self) -> Dict[str, Any]: : return {“param1”: 1, “param2”: 2} <br/> def forward(self, input: torch.Tensor, input_lengths: torch.Tensor) -> : Tuple[torch.Tensor, torch.Tensor]: # Your forward implementation here pass