espnet2.tasks.audio_metric.AudioMetricTask

source

class espnet2.tasks.audio_metric.AudioMetricTask

Bases: AbsTask

Train Uni-VERSA and autoregressive ARECHO audio metric predictors.

classmethod add_task_arguments(parser: ArgumentParser)

Register predictor, metadata, and preprocessing options.

classmethod build_collate_fn(args: Namespace, train: bool) → Callable[[Collection[Tuple[str, Dict[str, ndarray]]]], Tuple[List[str], Dict[str, Tensor]]]

Collate scalar metrics or autoregressive metric token sequences.

classmethod build_model(args: Namespace) → ESPnetUniversaModel

Build a predictor and embed its metadata in the saved configuration.

classmethod build_preprocess_fn(args: Namespace, train: bool) → Callable[[str, Dict[str, array]], Dict[str, ndarray]] | None

Prepare text and metric targets for the selected predictor.

class_choices_list : List[[ClassChoices](../train/ClassChoices.md#espnet2.train.class_choices.ClassChoices)] = [<espnet2.train.class_choices.ClassChoices object>, <espnet2.train.class_choices.ClassChoices object>]

num_optimizers : int = 1

classmethod optional_data_names(train: bool = True, inference: bool = False) → Tuple[str, ...]

Allow optional reference audio and text in every mode.

classmethod required_data_names(train: bool = True, inference: bool = False) → Tuple[str, ...]

Require metric targets only during training and validation.

trainer

alias of Trainer