Update state.py

state.py

@@ -723,6 +723,7 @@ def analyze(inFileName):
         _is_annotation_file = inFileName.lower().endswith((".rttm", ".txt", ".csv"))
         if _is_annotation_file:
             from pyannote.core import Annotation, Segment
+            from collections import defaultdict
             noVoice    = Annotation()
             multiVoice = Annotation()
             oneVoice   = Annotation()
@@ -733,14 +734,46 @@ def analyze(inFileName):
                  for seg in currAnnotation.subset([label]).itersegments()],
                 key=lambda x: x[0]
             )
-            prev_end = 0.0
+            # Detect multi-voice: pairwise overlaps between speakers
+            speaker_segs = defaultdict(list)
             for start, end, label in all_segs:
-                if start > prev_end + 0.1:
-                    noVoice[Segment(prev_end, start)] = 'silence'
-                oneVoice[Segment(start, end)] = label
-                prev_end = max(prev_end, end)
+                speaker_segs[label].append((start, end))
+            multi_intervals = []
+            labels_list = list(speaker_segs.keys())
+            for i in range(len(labels_list)):
+                for j in range(i+1, len(labels_list)):
+                    for s1, e1 in speaker_segs[labels_list[i]]:
+                        for s2, e2 in speaker_segs[labels_list[j]]:
+                            ov_s, ov_e = max(s1, s2), min(e1, e2)
+                            if ov_e > ov_s + 0.05:
+                                multi_intervals.append((ov_s, ov_e))
+            multi_intervals.sort()
+            merged_multi = []
+            for s, e in multi_intervals:
+                if merged_multi and s <= merged_multi[-1][1]:
+                    merged_multi[-1] = (merged_multi[-1][0], max(merged_multi[-1][1], e))
+                else:
+                    merged_multi.append([s, e])
+            for s, e in merged_multi:
+                multiVoice[Segment(s, e)] = 'overlap'
+            # No Voice: gaps in the union of all speech
+            speech_union = []
+            for start, end, _ in all_segs:
+                if speech_union and start <= speech_union[-1][1]:
+                    speech_union[-1] = (speech_union[-1][0], max(speech_union[-1][1], end))
+                else:
+                    speech_union.append([start, end])
+            prev_end = 0.0
+            for s, e in speech_union:
+                if s > prev_end + 0.1:
+                    noVoice[Segment(prev_end, s)] = 'silence'
+                prev_end = e
             if currTotalTime > prev_end + 0.1:
                 noVoice[Segment(prev_end, currTotalTime)] = 'silence'
+            # Single Voice: segments not overlapping any multi-voice region
+            for start, end, label in all_segs:
+                if not any(ms < end and me > start for ms, me in merged_multi):
+                    oneVoice[Segment(start, end)] = label
         else:
             try:
                 noVoice, oneVoice, multiVoice = su.calcSpeakingTypes(pipeline, currAnnotation, currTotalTime)