def _cer_update(
preds: Union[str, List[str]],
target: Union[str, List[str]],
) -> Tuple[Tensor, Tensor]:
"""Update the cer score with the current set of references and predictions.
Args:
preds: Transcription(s) to score as a string or list of strings
target: Reference(s) for each speech input as a string or list of strings
Returns:
Number of edit operations to get from the reference to the prediction, summed over all samples
Number of character overall references
"""
if isinstance(preds, str):
preds = [preds]
if isinstance(target, str):
target = [target]
errors = tensor(0, dtype=torch.float)
total = tensor(0, dtype=torch.float)
for pred, tgt in zip(preds, target):
pred_tokens = pred
tgt_tokens = tgt
errors += _edit_distance(list(pred_tokens), list(tgt_tokens))
total += len(tgt_tokens)
return errors, total
def _wip_update(
preds: Union[str, List[str]],
target: Union[str, List[str]],
) -> Tuple[Tensor, Tensor, Tensor]:
"""Update the wip score with the current set of references and predictions.
Args:
preds: Transcription(s) to score as a string or list of strings
target: Reference(s) for each speech input as a string or list of strings
Returns:
Number of edit operations to get from the reference to the prediction, summed over all samples
Number of words overall references
Number of words overall prediction
"""
if isinstance(preds, str):
preds = [preds]
if isinstance(target, str):
target = [target]
total = tensor(0.0)
errors = tensor(0.0)
target_total = tensor(0.0)
preds_total = tensor(0.0)
for pred, tgt in zip(preds, target):
pred_tokens = pred.split()
target_tokens = tgt.split()
errors += _edit_distance(pred_tokens, target_tokens)
target_total += len(target_tokens)
preds_total += len(pred_tokens)
total += max(len(target_tokens), len(pred_tokens))
return errors - total, target_total, preds_total