[Bugfix][CI] Vendor Qwen3-TTS and CosyVoice3 reference audio fixtures (#3267)

Signed-off-by: Yueqian Lin <linyueqian@outlook.com>

.gitattributes

@@ -27,6 +27,7 @@
 *.mov binary
 *.mp4 binary
 *.mp3 binary
+*.wav binary
 *.flv binary
 *.fla binary
 *.swf binary

.gitignore

@@ -264,6 +264,8 @@ tmp_test
 vllm_omni/_version.py
 # output files
 *.wav
+# Vendored test reference audio (small, intentionally checked in)
+!tests/assets/**/*.wav
 .worktrees/
 # CI overlay yamls materialized from tests/utils.py:_CI_OVERLAYS at test time
 tests/.ci_generated/

tests/e2e/offline_inference/test_cosyvoice3.py

@@ -13,10 +13,8 @@ verifies a stable reference recipe:
 from __future__ import annotations
 
 import functools
-import io
 import os
 from pathlib import Path
-from urllib.request import urlopen
 
 import numpy as np
 import pytest
@@ -25,6 +23,7 @@ from huggingface_hub import snapshot_download
 from vllm.sampling_params import SamplingParams
 
 from tests.helpers.mark import hardware_test
+from tests.helpers.media import test_asset_path
 from tests.helpers.runtime import OmniRunner
 from tests.helpers.stage_config import get_deploy_config_path
 from vllm_omni.model_executor.models.cosyvoice3.config import CosyVoice3Config
@@ -33,7 +32,9 @@ from vllm_omni.model_executor.models.cosyvoice3.tokenizer import get_qwen_tokeni
 MODEL = "FunAudioLLM/Fun-CosyVoice3-0.5B-2512"
 MODEL_DIR_ENV = "VLLM_OMNI_COSYVOICE3_MODEL_DIR"
 
-REFERENCE_PROMPT_WAV_URL = "https://raw.githubusercontent.com/FunAudioLLM/CosyVoice/main/asset/zero_shot_prompt.wav"
+# Vendored under tests/assets/cosyvoice3/ so the test does not depend on
+# raw.githubusercontent.com being reachable from CI runners.
+REFERENCE_PROMPT_WAV_PATH = test_asset_path("cosyvoice3/zero_shot_prompt.wav")
 REFERENCE_PROMPT_TEXT = "You are a helpful assistant.<|endofprompt|>希望你以后能够做的比我还好呦。"
 REFERENCE_SYNTH_TEXT = (
     "CosyVoice is undergoing a comprehensive upgrade, providing more accurate, "
@@ -51,9 +52,7 @@ ASYNC_CHUNK_MODES = [
 
 @functools.lru_cache(maxsize=1)
 def _load_reference_prompt_wav() -> tuple[np.ndarray, int]:
-    with urlopen(REFERENCE_PROMPT_WAV_URL, timeout=30) as resp:
-        data = resp.read()
-    audio, sr = sf.read(io.BytesIO(data), dtype="float32", always_2d=False)
+    audio, sr = sf.read(str(REFERENCE_PROMPT_WAV_PATH), dtype="float32", always_2d=False)
     if isinstance(audio, np.ndarray) and audio.ndim > 1:
         audio = np.mean(audio, axis=-1)
     return np.asarray(audio, dtype=np.float32), int(sr)

tests/e2e/offline_inference/test_qwen3_tts_base.py

@@ -16,10 +16,12 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 import pytest
 
 from tests.helpers.mark import hardware_test
+from tests.helpers.media import load_test_audio_data_url
 from tests.helpers.stage_config import get_deploy_config_path, modify_stage_config
 
 MODEL = "Qwen/Qwen3-TTS-12Hz-0.6B-Base"
-REF_AUDIO_URL = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone_2.wav"
+# See tests/e2e/online_serving/test_qwen3_tts_base.py for the vendored-asset rationale.
+REF_AUDIO_URL = load_test_audio_data_url("qwen3_tts/clone_2.wav")
 REF_TEXT = "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it! And thanks to you."
 
 

tests/e2e/online_serving/test_cosyvoice3_tts.py

@@ -15,13 +15,16 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 import pytest
 
 from tests.helpers.mark import hardware_test
+from tests.helpers.media import load_test_audio_data_url
 from tests.helpers.runtime import OmniServerParams
 from tests.helpers.stage_config import get_deploy_config_path
 
 MODEL = "FunAudioLLM/Fun-CosyVoice3-0.5B-2512"
 
-# Official CosyVoice zero-shot prompt audio and its transcript
-REF_AUDIO_URL = "https://raw.githubusercontent.com/FunAudioLLM/CosyVoice/main/asset/zero_shot_prompt.wav"
+# Official CosyVoice zero-shot prompt audio and its transcript. Vendored under
+# tests/assets/ so the server does not depend on raw.githubusercontent.com being
+# reachable at request time (same rationale as issue #3263 for Qwen3-TTS).
+REF_AUDIO_URL = load_test_audio_data_url("cosyvoice3/zero_shot_prompt.wav")
 REF_TEXT = "希望你以后能够做的比我还好呦。"
 
 

tests/e2e/online_serving/test_qwen3_tts_base.py

@@ -15,12 +15,16 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 import pytest
 
 from tests.helpers.mark import hardware_test
+from tests.helpers.media import load_test_audio_data_url
 from tests.helpers.runtime import OmniServerParams
 from tests.helpers.stage_config import get_deploy_config_path
 
 MODEL = "Qwen/Qwen3-TTS-12Hz-0.6B-Base"
 
-REF_AUDIO_URL = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone_2.wav"
+# Vendored under tests/assets/qwen3_tts/clone_2.wav so the server does not need
+# to fetch the reference audio over HTTPS at request time (CI runners in some
+# regions cannot reach the original Aliyun OSS host; see issue #3263).
+REF_AUDIO_URL = load_test_audio_data_url("qwen3_tts/clone_2.wav")
 REF_TEXT = "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it! And thanks to you."
 
 

tests/e2e/online_serving/test_qwen3_tts_base_expansion.py

@@ -12,6 +12,7 @@ import os
 import pytest
 
 from tests.helpers.mark import hardware_test
+from tests.helpers.media import load_test_audio_data_url
 from tests.helpers.runtime import OmniServerParams
 from tests.helpers.stage_config import get_deploy_config_path
 
@@ -22,7 +23,8 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "0"
 
 MODEL = "Qwen/Qwen3-TTS-12Hz-0.6B-Base"
 
-REF_AUDIO_URL = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-TTS-Repo/clone_2.wav"
+# See tests/e2e/online_serving/test_qwen3_tts_base.py for the vendored-asset rationale.
+REF_AUDIO_URL = load_test_audio_data_url("qwen3_tts/clone_2.wav")
 REF_TEXT = "Okay. Yeah. I resent you. I love you. I respect you. But you know what? You blew it! And thanks to you."
 
 

tests/helpers/media.py

@@ -496,6 +496,33 @@ def generate_synthetic_image(
     }
 
 
+_TEST_ASSETS_ROOT = Path(__file__).resolve().parents[1] / "assets"
+
+_AUDIO_MIME_BY_SUFFIX = {
+    ".wav": "audio/wav",
+    ".mp3": "audio/mpeg",
+    ".flac": "audio/flac",
+    ".ogg": "audio/ogg",
+}
+
+
+def test_asset_path(relative_path: str | os.PathLike) -> Path:
+    """Resolve a path under ``tests/assets/`` to its absolute on-disk location."""
+    return _TEST_ASSETS_ROOT / Path(relative_path)
+
+
+def load_test_audio_data_url(relative_path: str | os.PathLike) -> str:
+    """Load a vendored test audio file under ``tests/assets/`` as a base64 data URL.
+
+    Used by tests that need real reference audio (e.g. voice cloning) without
+    relying on the server's ability to fetch external URLs at request time.
+    """
+    path = test_asset_path(relative_path)
+    mime = _AUDIO_MIME_BY_SUFFIX.get(path.suffix.lower(), "application/octet-stream")
+    encoded = base64.b64encode(path.read_bytes()).decode("ascii")
+    return f"data:{mime};base64,{encoded}"
+
+
 def decode_b64_image(b64: str):
     img = Image.open(io.BytesIO(base64.b64decode(b64)))
     img.load()
@@ -653,5 +680,7 @@ __all__ = [
     "generate_synthetic_audio",
     "generate_synthetic_image",
     "generate_synthetic_video",
+    "load_test_audio_data_url",
     "preprocess_text",
+    "test_asset_path",
 ]