Update app/services/processor.py

app/services/processor.py

@@ -8,7 +8,6 @@ from collections import defaultdict, Counter
 
 import numpy as np
 import librosa
-import torch
 
 from app.core.config import get_settings
 from app.services.transcription import TranscriptionService
@@ -173,12 +172,11 @@ def merge_consecutive_segments(
 
         gap = seg.start - prev.end
 
-        overlap = prev.end - seg.start
 
         if (
             seg.speaker == prev.speaker
+            and gap >= -max_overlap
             and gap <= max_gap
-            and overlap <= max_overlap
         ):
 
             prev.end = max(
@@ -225,7 +223,6 @@ class Processor:
 
         # 2: Load audio
         y, sr = librosa.load(wav_path, sr=16000, mono=True)
-        waveform = torch.from_numpy(y).unsqueeze(0)
         if y.size == 0:
             raise ValueError("Empty audio")
         duration = len(y) / sr
@@ -248,8 +245,8 @@ class Processor:
 
         diarization_segments = [
             SpeakerSegment(
-                start=new_start,
-                end=new_end,
+                start=s.start,
+                end=s.end,
                 speaker=s.speaker,
                 confidence=getattr(s, "confidence", 1.0),
             )