tolerance should be set correctly in K2SpeechRecognitionDataset for importing kaldi data

[nuaalixu]

when converting data form old kaldi format to lhotse, it is recommand to use tolerance 100ms.
e.g. In lhotse/lhotse/bin/modes/kaldi.py

    CutSet.from_manifests(
        recordings=recording_set,
        supervisions=maybe_supervision_set,
        features=maybe_feature_set,
        tolerance=0.1,

However, It will raise an exception when the length of Supervisons and Features mismatch.

AssertionError: Supervisions ending after the cut are not supported for ASR 

The exception is due to the value of tol is fixed and too small:
In lhotse/lhotse/dataset/speech_recognition.py

def validate_for_asr(cuts: CutSet) -> None:
    validate(cuts)
    tol = 2e-3  # 1ms
    for cut in cuts:
        for supervision in cut.supervisions:
            assert supervision.start >= -tol, (
                f"Supervisions starting before the cut are not supported for ASR"
                f" (sup id: {supervision.id}, cut id: {cut.id})"
            )

            # Supervision start time is relative to Cut ...
            # https://lhotse.readthedocs.io/en/v0.10_e/cuts.html
            #
            # 'supervision.end' is end of supervision inside the Cut
            assert supervision.end <= cut.duration + tol, (
                f"Supervisions ending after the cut "
                f"are not supported for ASR"
                f" (sup id: {supervision.id}, cut id: {cut.id})"
            )

I suspect tol = 2e-3 doesn't meet the intended specifications.

[pzelasko]

Use lhotse.qa.fix_manifests on the imported supervisions + recordings.