Build the index for the speaker names. Args: hyp: a list of tuples, where each tuple is (speaker, start, end) of type (string,

build meta information from single rttm file path. Args: rttm_file: rttm file path Returns: rttm_dict: {'filename': {'spkname'

merge overlapped segments Args: segments: [(start, end), ...] Returns: segments: [(start, end), ...]

Build the connection matrix. Args: ref: a list of tuples for the ground truth, where each tuple is (speaker, start, end) of ty

Build the cost matrix. Args: ref: a list of tuples for the ground truth, where each tuple is (speaker, start, end) of type (st

Build the index for the speakers. Args: ref: a list of tuples for the ground truth, where each tuple is (speaker, start, end)

Compute the intersection length of two tuples. Args: A: a (speaker, start, end) tuple of type (string, float, float) B: a (speaker

Get all connected sub-graph in connection_matrix Args: connection_matrix: a 2-dim numpy array, whose element (i, j) = 1 means there exists

Get dynamic iou for segments with varying length because the fixed IoU (like 0.5) is too loose for long utterance Args: single_ref_segs: a

Calculate false alarm and missed speech duration Args: ref_segments: a sorted list with item (start, end) of type (float, float) h

Calculate the IoU of between segments_a listsand segments_b list Args: segments_a: a list with item (start, end) of type (float, float)

build meta information from ref and hyp rttm file path. Args: ref_rttm_file, hyp_rttm_file: rttm file path Returns: rttm_dict:

Get all nodes in a connected sub-graph via depth-first traversal Args: connection_matrix: a 2-dim numpy array, whose element (i, j) = 1 me