espnet3.systems.base.inference.InferenceRunner

source

class espnet3.systems.base.inference.InferenceRunner(provider: EnvironmentProvider, , async_mode: bool = False, async_specs_dir: str | Path = './_async_specs', async_result_dir: str | Path = './_async_results')

Bases: BaseRunner

Runner that produces hypotheses and references for each sample.

Example

>>> runner = InferenceRunner(provider)
>>> outputs = runner([0, 1])

static forward(idx, dataset=None, model=None, **kwargs)

Run inference for a single dataset index.

• Parameters:
  • idx (int) – Sample index to decode.
  • dataset – Dataset instance providing speech and text fields.
  • model – Inference model supporting callable interface and tokenizer.
  • **kwargs – Unused additional arguments.
• Returns: Mapping with idx, hyp (decoded text), and ref (reference text).
• Return type: Dict[str, Any]

Example

>>> InferenceRunner.forward(0, dataset=ds, model=asr_model)
{'idx': 0, 'hyp': '...', 'ref': '...'}