p_percentile_max=0.95,
init_search_step=0.01,
search_level=3)
icassp2018_clusterer = SpectralClusterer(
min_clusters=2,
max_clusters=18,
autotune=None,
laplacian_type=None,
refinement_options=icassp2018_refinement_options,
custom_dist="cosine")
for idx, sample_id in enumerate(sample_ids):
labels = icassp2018_clusterer.predict(sequences[idx])
print('Predicted labels: ', sample_id, f' {idx+1}/{len(sample_ids)}')
annotation = Annotation()
annotation.uri = sample_id
for jdx, speaker_id in enumerate(labels):
segment_interval = intervals[idx][jdx]
annotation[Segment(segment_interval[0],
segment_interval[1])] = speaker_id
rttm_file = '{}/{}.rttm'.format(rttm_dir, sample_id)
with open(rttm_file, 'w') as file:
annotation.support().write_rttm(file)
# rttm_file_collar = '{}/rttm_colar/{}.rttm'.format(rttm_dir, sample_id)
# with open(rttm_file_collar, 'w') as file:
# annotation.support(0.481).write_rttm(file)
