update timespoken w multi

state.py

@@ -715,24 +715,97 @@ def analyze(inFileName):
         printV("Loaded results", 4)
 
         pipeline = st.session_state.pipeline
-        try:
-            noVoice, oneVoice, multiVoice = su.calcSpeakingTypes(pipeline, currAnnotation, currTotalTime)
-        except Exception as e:
-            print(f"calcSpeakingTypes failed ({e}), falling back to annotation-based voice split")
-            # Fall back: build a clean annotation containing only properly named
-            # speakers (non-None, non-empty) so build_df5 never receives None labels.
-            from pyannote.core import Annotation
+        # For annotation-only files (RTTM/TXT/CSV), annotationToNoiseList
+        # misclassifies almost everything as multiVoice because the window-based
+        # classifier sees >2 speakers in every window. Instead derive voice
+        # categories directly from the annotation's segment gaps — more accurate
+        # and consistent for demo/pre-labeled files.
+        _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()
-            for label in currAnnotation.labels():
-                if label is not None and str(label).strip() != "":
-                    for seg in currAnnotation.subset([label]).itersegments():
-                        oneVoice[seg] = label
+            all_segs = sorted(
+                [(seg.start, seg.end, label)
+                 for label in currAnnotation.labels()
+                 if label is not None and str(label).strip() != ""
+                 for seg in currAnnotation.subset([label]).itersegments()],
+                key=lambda x: x[0]
+            )
+            # Detect multi-voice: pairwise overlaps between speakers
+            speaker_segs = defaultdict(list)
+            for start, end, label in all_segs:
+                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:
+                active = sorted({label for start, end, label in all_segs
+                                 if start < e and end > s})
+                mv_label = '+'.join(active) if active else 'overlap'
+                multiVoice[Segment(s, e)] = mv_label
+            # 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)
+            except Exception as e:
+                print(f"calcSpeakingTypes failed ({e}), falling back to annotation-based voice split")
+                from pyannote.core import Annotation, Segment
+                noVoice    = Annotation()
+                multiVoice = Annotation()
+                oneVoice   = Annotation()
+                all_segs = sorted(
+                    [(seg.start, seg.end, label)
+                     for label in currAnnotation.labels()
+                     if label is not None and str(label).strip() != ""
+                     for seg in currAnnotation.subset([label]).itersegments()],
+                    key=lambda x: x[0]
+                )
+                prev_end = 0.0
+                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)
+                if currTotalTime > prev_end + 0.1:
+                    noVoice[Segment(prev_end, currTotalTime)] = 'silence' 
         sumNoVoice    = su.sumTimes(noVoice)
         sumOneVoice   = su.sumTimes(oneVoice)
         sumMultiVoice = su.sumTimes(multiVoice)
 
+
         # df3
         df3 = utils.build_df3(noVoice, oneVoice, multiVoice)
         st.session_state.summaries[inFileName]["df3"] = df3
@@ -765,8 +838,6 @@ def analyze(inFileName):
             currTotalTime,
         )
         st.session_state.summaries[inFileName]["df2"] = df2
-
-        # Per-speaker multi-voice seconds — stored for the bar chart overlay
         mv_speakers, mv_times = su.sumMultiTimesPerSpeaker(multiVoice)
         st.session_state.summaries[inFileName]["mv_per_speaker"] = dict(zip(mv_speakers, mv_times))
         printV("Set df2", 4)