continue fixing

sentence_api/processing.py

@@ -9,6 +9,7 @@ from .audio_metrics import (
     analyze_samples,
     decode_audio_mono,
     refine_sentence_end_ms,
+    refine_sentence_start_ms,
 )
 from .config import Settings
 from .generate_boundaries import ALGORITHM_VERSION, make_entry
@@ -116,7 +117,8 @@ def document_from_transcript(
     previous_end = 0
     sentence_segments = split_sentences_at_punctuation(transcript.segments)
     for index, sentence_segment in enumerate(sentence_segments):
-        start_ms = max(previous_end, int(round(sentence_segment.start_seconds * 1000)))
+        raw_start_ms = int(round(sentence_segment.start_seconds * 1000))
+        start_ms = max(previous_end, raw_start_ms)
         raw_end_ms = int(round(sentence_segment.end_seconds * 1000))
         end_ms = min(
             duration_ms,
@@ -130,6 +132,14 @@ def document_from_transcript(
         end_ms = refine_sentence_end_ms(
             samples, sample_rate, raw_end_ms=raw_end_ms, padded_end_ms=end_ms
         )
+        refined_start_ms = refine_sentence_start_ms(
+            samples,
+            sample_rate,
+            raw_start_ms=raw_start_ms,
+            previous_end_ms=previous_end,
+        )
+        if previous_end <= refined_start_ms and end_ms - refined_start_ms >= 100:
+            start_ms = refined_start_ms
         if not sentence_segment.text.strip() or end_ms <= start_ms:
             continue
         start_sample = max(0, int(start_ms / 1000 * sample_rate))