diff --git a/app.py b/app.py
index 15448281..bd8688ff 100644
--- a/app.py
+++ b/app.py
@@ -1,3 +1,4 @@
+import gc
 import os
 import re
 import sys
@@ -5,6 +6,7 @@
 import random
 import numpy as np
 import gradio as gr
+import torch
 from typing import Optional, Tuple
 from funasr import AutoModel
 from pathlib import Path
@@ -265,12 +267,25 @@ def get_or_load_asr_model(self) -> AutoModel:
     def prompt_wav_recognition(self, prompt_wav: Optional[str]) -> str:
         if prompt_wav is None:
             return ""
-        res = self.get_or_load_asr_model().generate(
-            input=prompt_wav,
-            language="auto",
-            use_itn=True,
-        )
-        return res[0]["text"].split("|>")[-1]
+        try:
+            res = self.get_or_load_asr_model().generate(
+                input=prompt_wav,
+                language="auto",
+                use_itn=True,
+            )
+            return res[0]["text"].split("|>")[-1]
+        finally:
+            self.release_asr_model()
+
+    def release_asr_model(self) -> None:
+        """Release the request-scoped ASR model before TTS generation."""
+        if self.asr_model is None:
+            return
+        self.asr_model = None
+        gc.collect()
+        if self.asr_device.startswith("cuda") and torch.cuda.is_available():
+            torch.cuda.empty_cache()
+        logger.info("ASR model released after reference transcription.")
 
     def _build_generate_kwargs(
         self,
diff --git a/tests/test_app_asr_lifecycle.py b/tests/test_app_asr_lifecycle.py
new file mode 100644
index 00000000..9f2f1772
--- /dev/null
+++ b/tests/test_app_asr_lifecycle.py
@@ -0,0 +1,79 @@
+from __future__ import annotations
+
+import importlib.util
+import sys
+import types
+from pathlib import Path
+
+import pytest
+import torch
+
+
+ROOT = Path(__file__).resolve().parents[1]
+APP_PATH = ROOT / "app.py"
+
+
+class _I18n:
+    def __init__(self, **translations):
+        self.translations = translations
+
+    def __call__(self, key):
+        return key
+
+
+@pytest.fixture
+def app_module(monkeypatch):
+    gradio_stub = types.ModuleType("gradio")
+    gradio_stub.I18n = _I18n
+    gradio_stub.themes = types.SimpleNamespace(
+        Soft=lambda **kwargs: kwargs,
+        GoogleFont=lambda name: name,
+    )
+    monkeypatch.setitem(sys.modules, "gradio", gradio_stub)
+
+    funasr_stub = types.ModuleType("funasr")
+    funasr_stub.AutoModel = object
+    monkeypatch.setitem(sys.modules, "funasr", funasr_stub)
+
+    voxcpm_stub = types.ModuleType("voxcpm")
+    voxcpm_stub.VoxCPM = object
+    monkeypatch.setitem(sys.modules, "voxcpm", voxcpm_stub)
+    model_stub = types.ModuleType("voxcpm.model")
+    monkeypatch.setitem(sys.modules, "voxcpm.model", model_stub)
+    utils_stub = types.ModuleType("voxcpm.model.utils")
+    utils_stub.resolve_runtime_device = lambda requested, default: "cuda:0"
+    monkeypatch.setitem(sys.modules, "voxcpm.model.utils", utils_stub)
+
+    spec = importlib.util.spec_from_file_location("voxcpm_demo_app", APP_PATH)
+    module = importlib.util.module_from_spec(spec)
+    assert spec.loader is not None
+    spec.loader.exec_module(module)
+    return module
+
+
+@pytest.mark.parametrize("generate_error", [None, RuntimeError("ASR failed")])
+def test_prompt_recognition_releases_cuda_asr_model(app_module, monkeypatch, generate_error):
+    empty_cache_calls = []
+    monkeypatch.setattr(torch.cuda, "is_available", lambda: True)
+    monkeypatch.setattr(torch.cuda, "empty_cache", lambda: empty_cache_calls.append(True))
+
+    class _FakeAutoModel:
+        def __init__(self, **kwargs):
+            pass
+
+        def generate(self, **kwargs):
+            if generate_error is not None:
+                raise generate_error
+            return [{"text": "<|zh|>测试文本"}]
+
+    monkeypatch.setattr(app_module, "AutoModel", _FakeAutoModel)
+    demo = app_module.VoxCPMDemo()
+
+    if generate_error is None:
+        assert demo.prompt_wav_recognition("reference.wav") == "测试文本"
+    else:
+        with pytest.raises(RuntimeError, match="ASR failed"):
+            demo.prompt_wav_recognition("reference.wav")
+
+    assert demo.asr_model is None
+    assert empty_cache_calls == [True]