espnet3.systems.tts.system.TTSSystem

source

class espnet3.systems.tts.system.TTSSystem(training_config: DictConfig | None = None, inference_config: DictConfig | None = None, metrics_config: DictConfig | None = None, publication_config: DictConfig | None = None, stage_log_mapping: dict | None = None)

Bases: BaseSystem

TTS-specific system with local trainer customization.

Initialize the system with optional stage configs.

• Parameters:
  • training_config – Training configuration for data preparation, statistics collection, and model training.
  • inference_config – Inference configuration used by the infer stage.
  • metrics_config – Measurement configuration used by the measure stage.
  • publication_config – Publication configuration for pack_model and upload_model stages.
  • stage_log_mapping – Optional per-stage log directory overrides.

collect_stats(*args, **kwargs)

Run the collect_stats stage using the configured trainer.

Prepares the training runtime (directories, parallelism, seed), then delegates to the trainer’s collect_stats method. Positional and keyword stage arguments are rejected to avoid silent misconfiguration.

• Parameters:
  • *args – Must be empty. Passing any positional argument raises ValueError via _reject_stage_args.
  • **kwargs – Must be empty. Passing any keyword argument raises ValueError via _reject_stage_args.
• Returns: None
• Raises:ValueError – If any positional or keyword arguments are passed.

Notes

The normalize: null pattern from recipe configs is intentionally preserved — no normalization is applied during stats collection.

####### Examples

>>> from omegaconf import OmegaConf
>>> cfg = OmegaConf.create({"exp_dir": "/tmp/exp"})
>>> system = TTSSystem(training_config=cfg)
>>> system.collect_stats()  # runs stats collection end-to-end

train(*args, **kwargs)

Run the training stage using the configured trainer.

Prepares the runtime, optionally saves the ESPnet config, then calls trainer.fit with any keyword arguments drawn from training_config.fit. For GAN models, a GANTTSLightningTrainer is used automatically; for all other models, ESPnet3LightningTrainer is used.

• Parameters:
  • *args – Must be empty. Passing any positional argument raises ValueError via _reject_stage_args.
  • **kwargs – Must be empty. Passing any keyword argument raises ValueError via _reject_stage_args.
• Returns: None
• Raises:ValueError – If any positional or keyword arguments are passed.

Notes

training_config.fit is forwarded verbatim to trainer.fit. Common keys include max_epochs, ckpt_path, etc.

####### Examples

>>> from omegaconf import OmegaConf
>>> cfg = OmegaConf.create({
...     "exp_dir": "/tmp/exp",
...     "task": "tts",
...     "model": {"_target_": "my.Model"},
...     "fit": {"max_epochs": 10},
... })
>>> system = TTSSystem(training_config=cfg)
>>> system.train()  # trains the model for 10 epochs