Recheck narrowband TTS content with stronger ASR

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>

tests/e2e/online_serving/test_qwen3_tts_customvoice_expansion.py

@@ -148,6 +148,9 @@ def test_sample_rate_001(omni_server, online_client) -> None:
         "stream": False,
         "response_format": "wav",
         "sample_rate": 8000,
+        # Whisper-small can mishear the narrowband "kilohertz" pronunciation.
+        # Recheck the same WAV with stronger ASR, preserving the 0.9 gate.
+        "transcript_escalation_model": "large-v3",
         "task_type": "CustomVoice",
         "voice": "vivian",
     }