espnet3.systems.asr.system.ASRSystem

source

class espnet3.systems.asr.system.ASRSystem(, train_config: DictConfig | None = None, infer_config: DictConfig | None = None, metric_config: DictConfig | None = None, publish_config: DictConfig | None = None, demo_config: DictConfig | None = None, demo_config_path: Path | None = None)

Bases: BaseSystem

ASR-specific system.

This system adds: : - Tokenizer training inside train()

Initialize the system with optional stage configs.

create_dataset(*args, **kwargs)

Create datasets using the configured helper function.

The callable is resolved from train_config.create_dataset.func and invoked with the remaining configuration values.

• Raises:RuntimeError – If the configuration does not specify a function.

get_stage_log_dir(stage: str) → Path

Return stage-specific log directories when configured.

The ASR system routes logs to artifact directories when available: : - create_dataset: train_config.create_dataset.dataset_dir or train_config.dataset_dir or train_config.data_dir.

• train_tokenizer: train_config.tokenizer.save_path.
• collect_stats: train_config.stats_dir.
• train/publish: train_config.exp_dir.
• infer: infer_config.infer_dir.
• measure: metric_config.infer_dir or infer_config.infer_dir.

If none of the stage-specific paths are configured, it falls back to BaseSystem.get_stage_log_dir (train_config.exp_dir or <cwd>/logs).

• Parameters:stage (str) – Stage name being executed.
• Returns: Directory where the stage log should be placed.
• Return type: Path

pack_model(*args, **kwargs)

Pack model artifacts into an espnet3 bundle.

train(*args, **kwargs)

Train the model, training the tokenizer first if needed.

This stage checks for a cached tokenizer model and runs tokenizer training before delegating to the base training routine.

• Raises:RuntimeError – If train_config.dataset_dir is not set.

train_tokenizer(*args, **kwargs)

Train a SentencePiece tokenizer based on configured text.

The text builder configured in train_config.tokenizer.text_builder is used to generate training text, which is then saved and consumed by the SentencePiece trainer.

• Raises:RuntimeError – If required tokenizer config is missing or invalid.