espnet3.systems.base.scp_utils.load_scp_fields

source

espnet3.systems.base.scp_utils.load_scp_fields(decode_dir: Path, test_name: str, inputs: List[str] | Dict[str, str], file_suffix: str = '.scp') → Dict[str, List[str]]

Load and align SCP files into a field-wise dictionary for evaluation.

This function reads all required SCP files from the given test set, validates their consistency, and returns a dictionary with:

• “utt_id”: list of sorted utterance IDs
• <alias>: list of aligned values for each key (e.g., “ref”, “hyp”)

• Parameters:

  • decode_dir (Path) – Root directory of decode results.
  • test_name (str) – Subdirectory name of the test set (e.g., “test-clean”).
  • inputs (Union *[*List *[*str ] , Dict *[*str , str ] ]) – SCP keys to read.
    • List[str]: alias and filename are the same.
    • Dict[str, str]: alias → filename.
  • file_suffix (str , optional) – SCP file extension. Default: “.scp”

• Returns: { : “utt_id”: […], “ref”: […], “hyp”: […], …

  }

• Return type: Dict[str, List[str]]

• Raises:AssertionError – If SCP files are missing or utterance IDs are inconsistent.

Example

>>> load_scp_fields(
...     Path("decode"),
...     "test-other",
...     inputs={"ref": "text", "hyp": "hypothesis"},
... )
{
    "utt_id": ["utt1", "utt2"],
    "ref": ["the cat", "a dog"],
    "hyp": ["the bat", "a log"]
}

Notes

• Utterance IDs are sorted to ensure consistent alignment.
• Useful as direct input to AbsMetrics._call_().