def test_from_records(annotation):
# Check that we can reconstruct an annotation from the
# output of itertracks.
records = annotation.itertracks(yield_label=True)
actual = Annotation.from_records(records)
expected = annotation
assert actual == expected
def read_segments_file(segments_path):
"""Read speech/non-speech annotations from Kaldi segments file."""
columns = ['utterance_uri', 'recording_uri', 'onset', 'offset']
segs_df = pd.read_csv(
segments_path, header=None, sep=' ', names=columns)
annotations = {}
for recording_uri, segs in segs_df.groupby('recording_uri'):
records = [(Segment(seg.onset, seg.offset), '_', 'speech')
for seg in segs.itertuples(index=False)]
ann = Annotation.from_records(records, uri=recording_uri)
annotations[recording_uri] = ann
return annotations