Bound transcript memory and translation queue size

- Add MAX_TRANSCRIPT_LENGTH (500) and MAX_TRANSLATION_QUEUE_SIZE (100)
  class constants to ServeClientBase
- Trim transcript and text lists after each update_segments() call
- Create translation queue with maxsize to prevent unbounded growth
- Added tests for _trim_transcript()

tests/test_base_backend.py

@@ -266,6 +266,33 @@ class TestCleanup(unittest.TestCase):
         self.assertTrue(client.exit)
 
 
+class TestTrimTranscript(unittest.TestCase):
+    def setUp(self):
+        self.ws = MagicMock()
+        self.client = ConcreteServeClient(client_uid="test", websocket=self.ws)
+
+    def test_transcript_trimmed_when_over_max(self):
+        self.client.transcript = [
+            {"start": f"{i}.000", "end": f"{i+1}.000", "text": f"seg{i}", "completed": True}
+            for i in range(self.client.MAX_TRANSCRIPT_LENGTH + 100)
+        ]
+        self.client._trim_transcript()
+        self.assertEqual(len(self.client.transcript), self.client.MAX_TRANSCRIPT_LENGTH)
+        self.assertEqual(self.client.transcript[0]["text"], "seg100")
+
+    def test_transcript_not_trimmed_when_under_max(self):
+        self.client.transcript = [
+            {"start": "0.000", "end": "1.000", "text": "a", "completed": True}
+        ]
+        self.client._trim_transcript()
+        self.assertEqual(len(self.client.transcript), 1)
+
+    def test_text_list_trimmed(self):
+        self.client.text = ["word"] * (self.client.MAX_TRANSCRIPT_LENGTH + 50)
+        self.client._trim_transcript()
+        self.assertEqual(len(self.client.text), self.client.MAX_TRANSCRIPT_LENGTH)
+
+
 class TestUpdateSegments(unittest.TestCase):
     """Tests for the core update_segments() logic."""
 

whisper_live/backend/base.py

@@ -24,6 +24,9 @@ class ServeClientBase(object):
     same_output_threshold: int
     """Number of repeated outputs before considering it as a valid segment."""
 
+    MAX_TRANSCRIPT_LENGTH = 500
+    MAX_TRANSLATION_QUEUE_SIZE = 100
+
     def __init__(
         self,
         client_uid,
@@ -376,4 +379,12 @@ class ServeClientBase(object):
             with self.lock:
                 self.timestamp_offset += offset
 
+        self._trim_transcript()
         return last_segment
+
+    def _trim_transcript(self):
+        """Trims transcript and text lists to prevent unbounded memory growth."""
+        if len(self.transcript) > self.MAX_TRANSCRIPT_LENGTH:
+            self.transcript = self.transcript[-self.MAX_TRANSCRIPT_LENGTH:]
+        if len(self.text) > self.MAX_TRANSCRIPT_LENGTH:
+            self.text = self.text[-self.MAX_TRANSCRIPT_LENGTH:]

whisper_live/server.py

@@ -191,7 +191,7 @@ class TranscriptionServer:
         
         if enable_translation:
             target_language = options.get("target_language", "fr")
-            translation_queue = queue.Queue()
+            translation_queue = queue.Queue(maxsize=ServeClientBase.MAX_TRANSLATION_QUEUE_SIZE)
             from whisper_live.backend.translation_backend import ServeClientTranslation
             translation_client = ServeClientTranslation(
                 client_uid=options["uid"],