diff --git a/tests/entrypoints/openai_api/test_serving_speech.py b/tests/entrypoints/openai_api/test_serving_speech.py
index 0c2006536..bed775ee2 100644
--- a/tests/entrypoints/openai_api/test_serving_speech.py
+++ b/tests/entrypoints/openai_api/test_serving_speech.py
@@ -39,6 +39,7 @@ from vllm_omni.entrypoints.openai.serving_speech import (
     OmniOpenAIServingSpeech,
     _create_wav_header,
 )
+from vllm_omni.entrypoints.openai.tts_adapters.base import PreparedRequest
 from vllm_omni.model_executor.models.fish_speech.prompt_utils import (
     FISH_TEXT_ONLY_SYSTEM_PROMPT,
     build_fish_voice_clone_prompt_ids,
@@ -3343,6 +3344,40 @@ class TestTTSAsyncOffloading:
         qwen3_tts_server._build_tts_params.assert_called_once()
         qwen3_tts_server._estimate_prompt_len_async.assert_awaited_once()
 
+    def test_prepare_speech_generation_uses_adapter_model_type_label(
+        self,
+        voxtral_server,
+        mocker: MockerFixture,
+    ):
+        """Adapter model_type should replace the legacy _tts_model_type label ladder."""
+        legacy_tts_model_type = "dummy_tts"
+        adapter_model_type = "adapter_dummy_tts"
+
+        class FakeAdapter:
+            def validate(self, request):
+                return None
+
+            async def build(self, request, sampling_params_list, has_inline_ref_audio):
+                return PreparedRequest(
+                    prompt={"prompt": request.input},
+                    tts_params={},
+                    model_type=adapter_model_type,
+                )
+
+        voxtral_server._tts_model_type = legacy_tts_model_type
+        mocker.patch.object(voxtral_server, "_get_tts_adapter", return_value=FakeAdapter())
+        log_info = mocker.patch("vllm_omni.entrypoints.openai.serving_speech.logger.info")
+
+        asyncio.run(voxtral_server._prepare_speech_generation(OpenAICreateSpeechRequest(input="hello")))
+
+        assert adapter_model_type != legacy_tts_model_type
+        assert any(
+            call.args
+            and call.args[0] == "TTS speech request %s: text=%r, model=%s"
+            and call.args[3] == adapter_model_type
+            for call in log_info.call_args_list
+        )
+
     def test_prepare_speech_generation_treats_sse_as_streaming(self, qwen3_tts_server, mocker: MockerFixture):
         """stream_format=sse should request delta-style multimodal outputs."""
         qwen3_tts_server._validate_tts_request = mocker.MagicMock(return_value=None)
diff --git a/vllm_omni/entrypoints/openai/serving_speech.py b/vllm_omni/entrypoints/openai/serving_speech.py
index 72654738f..6f8f1e6dc 100644
--- a/vllm_omni/entrypoints/openai/serving_speech.py
+++ b/vllm_omni/entrypoints/openai/serving_speech.py
@@ -3513,11 +3513,13 @@ class OmniOpenAIServingSpeech(OpenAIServing, AudioMixin):
         # apply uploaded speakers inside validate(), which sets request.ref_audio
         # in place. The builders need to know whether the caller supplied audio
         # inline vs. via an uploaded voice.
+        model_type: str | None = None
         has_inline_ref_audio = request.ref_audio is not None
         if self._tts_model_type == "ming_flash_omni_tts":
             # ming_flash_omni is intentionally NOT migrated onto the adapter
             # framework in this PR (it has no registered adapter); keep it on the
             # legacy inline dispatch so serving still works.
+            model_type = "ming_flash_omni_tts"
             validation_error = self._validate_ming_flash_omni_tts_request(request)
             if validation_error:
                 raise ValueError(validation_error)
@@ -3531,6 +3533,7 @@ class OmniOpenAIServingSpeech(OpenAIServing, AudioMixin):
             prepared = await adapter.build(request, sampling_params_list, has_inline_ref_audio)
             prompt = prepared.prompt
             tts_params = prepared.tts_params
+            model_type = prepared.model_type
             qwen3_ref_audio_warmup_artifact_key = prepared.warmup_artifact_key
         else:
             # Qwen omni models (Qwen3-Omni, Qwen2.5-Omni) use a "talker"
@@ -3554,36 +3557,11 @@ class OmniOpenAIServingSpeech(OpenAIServing, AudioMixin):
             tts_params = {}
             prompt = {"prompt": request.input}
 
-        if self._tts_model_type == "step_audio2":
-            model_type = "step_audio2"
-        elif self._is_fish_speech:
-            model_type = "fish_speech"
-        elif self._tts_model_type == "covo_audio":
-            model_type = "covo_audio"
-        elif self._tts_model_type == "voxtral_tts":
-            model_type = "voxtral_tts"
-        elif self._tts_model_type == "cosyvoice3":
-            model_type = "cosyvoice3"
-        elif self._tts_model_type == "voxcpm2":
-            model_type = "voxcpm2"
-        elif self._tts_model_type == "ming_flash_omni_tts":
-            model_type = "ming_flash_omni_tts"
-        elif self._tts_model_type == "ming_tts":
-            model_type = "ming_tts"
-        elif self._tts_model_type == "moss_tts_nano":
-            model_type = "moss_tts_nano"
-        elif self._tts_model_type == "moss_tts":
-            model_type = "moss_tts"
-        elif self._tts_model_type == "higgs_audio_v2":
-            model_type = "higgs_audio_v2"
-        elif self._tts_model_type == "glm_tts":
-            model_type = "glm_tts"
-        elif self._tts_model_type == "indextts2":
-            model_type = "indextts2"
-        elif self._is_tts:
-            model_type = tts_params.get("task_type", ["unknown"])[0]
-        else:
-            model_type = "generic"
+        if model_type is None:
+            if self._is_tts:
+                model_type = tts_params.get("task_type", ["unknown"])[0]
+            else:
+                model_type = "generic"
         logger.info(
             "TTS speech request %s: text=%r, model=%s",
             request_id,