espnet3.demo.runtime.run_inference

source

espnet3.demo.runtime.run_inference(runtime: DemoRuntime, , ui_names: List[str], ui_values: List[Any], output_names: List[str]) → List[Any]

Run a single inference pass and map outputs into UI order.

The UI inputs are received as parallel lists (names and values). These are normalized and wrapped into a minimal dataset, then executed via:

• runtime.runner_cls.forward when a runner is configured, or
• direct callable runtime.model when no runner is configured.

The raw output is then mapped to the UI outputs using runtime.output_keys.

• Parameters:
  • runtime (DemoRuntime) – Runtime created by build_runtime().
  • ui_names (List *[*str ]) – UI input field names (keys).
  • ui_values (List *[*Any ]) – UI input values, aligned with ui_names.
  • output_names (List *[*str ]) – UI output component names, used to order results.
• Returns: Output values aligned with output_names.
• Return type: List[Any]
• Raises:
  • ValueError – If UI output names and configured output_keys mismatch, or if no runner is configured and the input schema is unsupported.
  • RuntimeError – If a runner/model is not configured to execute inference.

Example

>>> outputs = run_inference(
...     runtime,
...     ui_names=["speech"],
...     ui_values=[waveform],
...     output_names=["text"],
... )

Notes

• Gradio audio inputs are normalized from (sample_rate, np.ndarray) to a float32 waveform NumPy array.