✨ Don't recombine paragraphs which only contain non-word text

e.g. [Music] or *Applause*

worker/transcribee_worker/whisper_transcribe.py

@@ -22,6 +22,11 @@
         r".*\d\.\s?$"
     ),  # Don't split on numerals followed by a dot, e.g. "during the 20. century"
 ]
+# Regexes that protect a paragraph from being recombined
+DONT_COMBINE_RES = [
+    re.compile(r"^\[[^\s]*\]$"),  # [MUSIC]
+    re.compile(r"^\*[^\s]*\*$"),  # *Applause*
+]
 
 
 def get_model_file(model_name: str):
@@ -262,6 +267,13 @@ async def strict_sentence_paragraphs(
                 lang=paragraph.lang, speaker=paragraph.speaker, children=[]
             )
 
+        elif any(regex.search(paragraph.text()) for regex in DONT_COMBINE_RES):
+            if acc_paragraph.children:
+                yield acc_paragraph
+            acc_paragraph = None
+            yield paragraph
+            continue
+
         locale = Locale(paragraph.lang)
         sentence_iter = BreakIterator.createSentenceInstance(locale)
         sentence_iter.setText(acc_paragraph.text() + paragraph.text())