diff --git a/test/hifi_clone.wav b/test/hifi_clone.wav
deleted file mode 100644
index c523afe..0000000
Binary files a/test/hifi_clone.wav and /dev/null differ
diff --git a/test/lzl.mp3 b/test/lzl.mp3
deleted file mode 100644
index a7c0a88..0000000
Binary files a/test/lzl.mp3 and /dev/null differ
diff --git a/test/lzl.wav b/test/lzl.wav
deleted file mode 100644
index 45d8073..0000000
Binary files a/test/lzl.wav and /dev/null differ
diff --git a/test/lzl_prompt_cache.pt b/test/lzl_prompt_cache.pt
deleted file mode 100644
index 31b786a..0000000
Binary files a/test/lzl_prompt_cache.pt and /dev/null differ
diff --git a/test/voice_design.wav b/test/voice_design.wav
deleted file mode 100644
index 4213f57..0000000
Binary files a/test/voice_design.wav and /dev/null differ
diff --git a/test/voxcpm_test.py b/test/voxcpm_test.py
deleted file mode 100644
index d2f8c94..0000000
--- a/test/voxcpm_test.py
+++ /dev/null
@@ -1,67 +0,0 @@
-from voxcpm import VoxCPM
-import soundfile as sf
-import torch
-from pathlib import Path
-import sys
-
-BASE_DIR = Path(__file__).resolve().parent
-DEVICE = "cuda"
-PROMPT_WAV = BASE_DIR / "lzl.wav"
-CACHE_PATH = BASE_DIR / "lzl_prompt_cache.pt"
-VOICE_DESIGN_PATH = BASE_DIR / "voice_design.wav"
-HIFI_CLONE_PATH = BASE_DIR / "hifi_clone.wav"
-PROMPT_TEXT = "亲爱的，今天的你要出发，挣钱喽，都没有系好安全带。 乖乖仔，我要到了，奇怪还没有分开就开始想你了，注意一下，闯红灯拍照，我可不喜欢明知孤犯的小坏蛋，安全带系一好，我们这边要出发喽。 小坏蛋，前方有限速拍照，姐姐给你盯着呢，车速太快了啊，慢一点慢一点降下来，不要让我害怕好吗？ 过最堵的路段，千万不要着急，姐姐会一直陪着你。 今日导航就先到这里了，哥哥注意安全停车哦。"
-TARGET_TEXT = "慢慢来吧，额度还在就好～"
-
-if DEVICE != "cuda":
-    raise RuntimeError(f"Unsupported device: {DEVICE}. This script only supports CUDA.")
-
-if not torch.cuda.is_available():
-    raise RuntimeError(
-        "This script requires CUDA. "
-        f"Current python: {sys.executable}, torch: {torch.__version__}. "
-        "Please run it with a CUDA-enabled Python environment."
-    )
-
-model = VoxCPM.from_pretrained("openbmb/VoxCPM2", load_denoiser=False)
-if model.tts_model.device != DEVICE:
-    raise RuntimeError(f"Expected VoxCPM to run on {DEVICE}, got {model.tts_model.device}")
-print(f"VoxCPM loaded on {model.tts_model.device} with torch {torch.__version__}")
-
-wav = model.generate(
-    text="(A young woman, gentle and sweet voice)Hello, welcome to VoxCPM!",
-    cfg_value=2.0,
-    inference_timesteps=10,
-)
-sf.write(str(VOICE_DESIGN_PATH), wav, model.tts_model.sample_rate)
-print(f"Saved voice design to {VOICE_DESIGN_PATH}")
-
-# Build and persist the prompt cache so later runs can skip prompt encoding.
-if CACHE_PATH.exists():
-    prompt_cache = torch.load(CACHE_PATH, map_location="cpu")
-    print(f"Loaded prompt cache from {CACHE_PATH}")
-else:
-    prompt_cache = model.tts_model.build_prompt_cache(
-        prompt_wav_path=str(PROMPT_WAV),
-        prompt_text=PROMPT_TEXT,
-        reference_wav_path=str(PROMPT_WAV),
-    )
-    torch.save(prompt_cache, CACHE_PATH)
-    print(f"Built and saved prompt cache to {CACHE_PATH}")
-
-cache_devices = {
-    key: str(value.device)
-    for key, value in prompt_cache.items()
-    if isinstance(value, torch.Tensor)
-}
-print(f"Prompt cache tensor devices: {cache_devices}")
-
-wav, _, _ = model.tts_model.generate_with_prompt_cache(
-    target_text=TARGET_TEXT,
-    prompt_cache=prompt_cache,
-    cfg_value=2.0,
-    inference_timesteps=10,
-)
-wav = wav.detach().cpu().float().squeeze(0).numpy()
-sf.write(str(HIFI_CLONE_PATH), wav, model.tts_model.sample_rate)
-print(f"Saved cloned audio to {HIFI_CLONE_PATH}")