[BugFix] Fix local media access for diffusion speech (#6622)

Signed-off-by: zhuhh97 <zhuhonghan@huawei.com>
This commit is contained in:
Honghan Zhu authored and GitHub committed 2026-08-29 02:01:24 +00:00
1 parent 034af1f657
commit f584729f24
4 files changed
+121 -15

No files matched your search

@@ -1245,6 +1245,46 @@ class TestTTSMethods:
cache_key
) == speech_server._make_ref_audio_artifact_cache_key(np.asarray(first[0], dtype=np.float32), 24000)
@pytest.mark.asyncio
async def test_diffusion_ref_audio_uses_media_access_config(
self,
mocker: MockerFixture,
monkeypatch,
tmp_path,
):
monkeypatch.setenv("SPEAKER_SAMPLES_DIR", str(tmp_path / "speakers"))
connector = mocker.MagicMock()
connector.fetch_audio_async = mocker.AsyncMock(return_value=(np.zeros(24000, dtype=np.float32), 24000))
connector_cls = mocker.patch.object(
serving_speech_module,
"MediaConnector",
return_value=connector,
)
media_dir = tmp_path / "media"
media_dir.mkdir()
ref_audio = media_dir / "reference.wav"
ref_audio.write_bytes(b"v1")
server = OmniOpenAIServingSpeech.for_diffusion(
diffusion_engine=mocker.MagicMock(),
model_name="test-model",
allowed_local_media_path=str(media_dir),
allowed_media_domains=["media.example.com"],
)
try:
first = await server._resolve_ref_audio(ref_audio.as_uri())
ref_audio.write_bytes(b"v2-with-different-size")
second = await server._resolve_ref_audio(ref_audio.as_uri())
finally:
server.shutdown()
connector_cls.assert_called_once_with(
allowed_local_media_path=str(media_dir),
allowed_media_domains=["media.example.com"],
)
assert first[2] != second[2]
assert connector.fetch_audio_async.await_count == 2
# ── ref-audio cache key tests (static helper) ──
def test_local_file_cache_key_invalidation(self):