Transform

class src.data.TransformManager.TransformManager(representation_name: str, transform_name: list, tensor_noisyspeech_transform_info_dir: str, tensor_cleanspeech_transform_info_dir: str)

Bases: object

Class that handles the transformation of the input data.

get_transform_info(tensor_transform_info_dir: str, map_location: str | None = None, verbose: bool = False) → dict

Method that loads transformation metadata.

Parameters:

• tensor_transform_info_dir (dict) -- transformation metadata directory.

• map_location (Optional[str]) -- Remap storage location.

• verbose (bool) -- Boolean that indicates weather to print specific output.

inverse_log(coefficients_transform: Tensor) → Tensor

inverse_log10(coefficients_transform: Tensor) → Tensor

inverse_log_power(coefficients_transform: Tensor) → Tensor

inverse_maxabs(coefficients_transform: Tensor) → Tensor

inverse_normalize(coefficients_transform: Tensor, low: float = 0.0, up: float = 1.0, epsilon: float = 1e-08, quantile_flag: bool = False) → Tensor

inverse_quantile_maxabs(coefficients_transform: Tensor) → Tensor

inverse_shift(coefficients_transform: Tensor) → Tensor

inverse_standardize(coefficients_transform: Tensor, mean_: float = 0.0, std_: float = 1.0) → Tensor

log(coefficients: Tensor, epsilon: float = 1e-08) → Tensor

Method that computes logarithmic power of input data.

Parameters:

• coefficients (torch.Tensor) -- Input data.

• epsilon (float) -- A small value to avoid computation error.

log10(coefficients: Tensor, epsilon: float = 1e-08) → Tensor

Method that computes loga10 of input data.

Parameters:

• coefficients (torch.Tensor) -- Input data.

• epsilon (float) -- A small value to avoid computation error.

log_power(coefficients: Tensor, epsilon: float = 1e-08) → Tensor

Method that computes logarithmic power of input data.

Parameters:

• coefficients (torch.Tensor) -- Input data.

• epsilon (float) -- A small value to avoid computation error.

maxabs(coefficients: Tensor) → Tensor

Method that scales input data by its maximum absolute value.

Parameters:

coefficients (torch.Tensor) -- Input data.

normalize(coefficients: Tensor, low: float = 0.0, up: float = 1.0, epsilon: float = 1e-08, quantile_flag: bool = False) → Tensor

Method that normalizes input data.

Parameters:

• coefficients (torch.Tensor) -- Input data.

• low (float) -- Lower bound.

• up (float) -- Upper bound.

• epsilon (float) -- A small value to avoid computation error.

• quantile_flag (bool) -- Boolean that indicates weather to scale data by its q-th quantile metadata.

quantile_maxabs(coefficients: Tensor) → Tensor

Method that scales input data by its q-th quantile metadata.

Parameters:

coefficients (torch.Tensor) -- Input data.

shift(coefficients: Tensor) → Tensor

Method that shifts input data.

Parameters:

coefficients (torch.Tensor) -- Input data.

standardize(coefficients: Tensor, mean_: float = 0.0, std_: float = 1.0) → Tensor

Method that standardizes input data.

Parameters:

• coefficients (torch.Tensor) -- Input data.

• mean (float) -- Target mean.

• std (float) -- Target standard deviation.