espnet2.tasks.ps2st.PS2STTask

source

class espnet2.tasks.ps2st.PS2STTask

Bases: AbsTask

PS2ST refers to the prompt-based speech-to-speech/text task.

The prompt is a text that serves as an instruction for the model to do a specific task such as ASR, IC, ST, etc. The output can be a text sequence or speech, depending on the task. For example, transcriptions for ASR, textual labels for classification, or synthesized speech for speech generation tasks.

classmethod add_task_arguments(parser: ArgumentParser)

Add task-specific arguments

classmethod build_collate_fn(args: Namespace, train: bool) → Callable

Build collate function

classmethod build_model(args: Namespace) → ESPnetQwen2AudioModel

Build the Qwen2-Audio model

classmethod build_preprocess_fn(args: Namespace, train: bool) → Callable | None

Build preprocessing function

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

num_optimizers : int = 1

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

Define optional data names

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

Define required data names

trainer

alias of Trainer