diff --git a/nemo/collections/asr/models/aed_multitask_models.py b/nemo/collections/asr/models/aed_multitask_models.py
index 91e5dd2b60c0..2de734600a18 100644
--- a/nemo/collections/asr/models/aed_multitask_models.py
+++ b/nemo/collections/asr/models/aed_multitask_models.py
@@ -569,7 +569,8 @@ def transcribe(
                     f"but got {type(override_config)}"
                 )
             trcfg = override_config
-            trcfg.timestamps = timestamps
+            if timestamps is not None:
+                trcfg.timestamps = timestamps
 
         if trcfg.enable_chunking:
             # Check if only one audio is provided with string
diff --git a/tests/collections/asr/mixins/test_transcription.py b/tests/collections/asr/mixins/test_transcription.py
index d2df529d9778..3c4cf7e2a38a 100644
--- a/tests/collections/asr/mixins/test_transcription.py
+++ b/tests/collections/asr/mixins/test_transcription.py
@@ -579,3 +579,30 @@ def test_transcribe_returns_xattn(self, audio_files, canary_1b_v2):
 
         # Reset the decoding strategy to original
         canary_1b_v2.change_decoding_strategy(orig_decoding_config)
+
+    @pytest.mark.with_downloads()
+    @pytest.mark.unit
+    def test_transcribe_override_config_preserves_timestamps(self, audio_files, canary_1b_v2):
+        canary_1b_v2.eval()
+        audio1, audio2 = audio_files
+
+        config = MultiTaskTranscriptionConfig(
+            batch_size=4,
+            return_hypotheses=True,
+            num_workers=0,
+            verbose=False,
+            prompt={'source_lang': 'en', 'target_lang': 'en'},
+            enable_chunking=False,
+            timestamps=True,
+        )
+
+        output = canary_1b_v2.transcribe([audio1, audio2], override_config=config)
+
+        assert len(output) == 2
+        assert isinstance(output[0], Hypothesis)
+        assert isinstance(output[1], Hypothesis)
+
+        assert output[0].timestamp is not None
+        assert output[1].timestamp is not None
+        assert 'word' in output[0].timestamp
+        assert 'word' in output[1].timestamp