Stub optional streaming test deps on Windows

backend/tests/unit/test_dg_start_guard.py

@@ -7,7 +7,7 @@
 import os
 import sys
 from types import ModuleType
-from unittest.mock import MagicMock, patch
+from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
 
@@ -49,6 +49,12 @@
 # MagicMock auto-generates attributes on access, and overwriting would pollute
 # shared pytest state for test_streaming_deepgram_backoff.py's close/error handler tests.
 
+_speaker_embedding = ModuleType('utils.stt.speaker_embedding')
+_speaker_embedding.SPEAKER_MATCH_THRESHOLD = 0.45
+_speaker_embedding.async_extract_embedding_from_bytes = AsyncMock(return_value=None)
+_speaker_embedding.compare_embeddings = MagicMock(return_value=0.0)
+sys.modules.setdefault('utils.stt.speaker_embedding', _speaker_embedding)
+
 # Now import the real streaming module
 from utils.stt.streaming import connect_to_deepgram
 

backend/tests/unit/test_parakeet_diarization.py

@@ -7,14 +7,56 @@
 
 import asyncio
 import os
+import sys
+from types import ModuleType
+from unittest.mock import AsyncMock, MagicMock
 
 import numpy as np
 
 os.environ.setdefault('HOSTED_SPEAKER_EMBEDDING_API_URL', 'http://fake')  # enables _diarize
 os.environ.setdefault('DEEPGRAM_API_KEY', 'x')
 
+_owned_modules = set()
+for _mod_name in [
+    'deepgram',
+    'deepgram.clients',
+    'deepgram.clients.live',
+    'deepgram.clients.live.v1',
+    'websockets',
+    'websockets.exceptions',
+]:
+    if _mod_name not in sys.modules:
+        sys.modules[_mod_name] = MagicMock()
+        _owned_modules.add(_mod_name)
+
+if 'deepgram' in _owned_modules:
+    sys.modules['deepgram'].DeepgramClient = MagicMock
+    sys.modules['deepgram'].DeepgramClientOptions = MagicMock
+    sys.modules['deepgram'].LiveTranscriptionEvents = MagicMock()
+if 'deepgram.clients.live.v1' in _owned_modules:
+    sys.modules['deepgram.clients.live.v1'].LiveOptions = MagicMock
+
+_speaker_embedding = ModuleType('utils.stt.speaker_embedding')
+_speaker_embedding.SPEAKER_MATCH_THRESHOLD = 0.45
+_speaker_embedding.async_extract_embedding_from_bytes = AsyncMock(return_value=None)
+
+
+def _cosine_distance(a, b):
+    a = np.asarray(a, dtype=np.float32)
+    b = np.asarray(b, dtype=np.float32)
+    denom = np.linalg.norm(a) * np.linalg.norm(b)
+    if denom == 0:
+        return 1.0
+    return float(1.0 - np.sum(a * b) / denom)
+
+
+_speaker_embedding.compare_embeddings = _cosine_distance
+sys.modules.setdefault('utils.stt.speaker_embedding', _speaker_embedding)
+
 import utils.stt.streaming as st  # noqa: E402
 
+st.compare_embeddings = _cosine_distance
+
 
 def _dir_vec(idx: int, rng) -> np.ndarray:
     """A unit direction in dim `idx` plus small within-speaker noise -> (1, 256)."""

backend/tests/unit/test_parakeet_stream_session.py

@@ -29,6 +29,50 @@
 mock_transcribe.INFERENCE_MODE = "nemo"
 sys.modules['transcribe'] = mock_transcribe
 
+_scipy = types.ModuleType('scipy')
+_scipy_spatial = types.ModuleType('scipy.spatial')
+_scipy_distance = types.ModuleType('scipy.spatial.distance')
+
+
+def _cosine_cdist(a, b, metric="cosine"):
+    if metric != "cosine":
+        raise ValueError(f"unsupported metric: {metric}")
+    a = np.asarray(a, dtype=np.float32)
+    b = np.asarray(b, dtype=np.float32)
+    a_norm = np.linalg.norm(a, axis=1, keepdims=True)
+    b_norm = np.linalg.norm(b, axis=1, keepdims=True).T
+    denom = a_norm * b_norm
+    similarity = np.divide(a @ b.T, denom, out=np.zeros((a.shape[0], b.shape[0]), dtype=np.float32), where=denom != 0)
+    return 1.0 - similarity
+
+
+_scipy_distance.cdist = _cosine_cdist
+_scipy_spatial.distance = _scipy_distance
+_scipy.spatial = _scipy_spatial
+sys.modules.setdefault('scipy', _scipy)
+sys.modules.setdefault('scipy.spatial', _scipy_spatial)
+sys.modules.setdefault('scipy.spatial.distance', _scipy_distance)
+
+_torch = types.ModuleType('torch')
+_torch.int16 = np.int16
+
+
+class _TorchArray:
+    def __init__(self, value):
+        self.value = np.asarray(value)
+
+    def float(self):
+        return self
+
+    def __truediv__(self, value):
+        return _TorchArray(self.value / value)
+
+
+_torch.frombuffer = lambda buffer, dtype: _TorchArray(np.frombuffer(buffer, dtype=dtype))
+_torch.hub = MagicMock()
+_torch.hub.load.side_effect = RuntimeError("torch hub unavailable in unit tests")
+sys.modules.setdefault('torch', _torch)
+
 import stream_handler as sh
 
 

backend/tests/unit/test_streaming_deepgram_backoff.py

@@ -8,6 +8,7 @@
 
 import asyncio
 import sys
+from types import ModuleType
 from unittest.mock import MagicMock, patch, AsyncMock
 
 import pytest
@@ -39,6 +40,19 @@
     sys.modules['deepgram'].LiveTranscriptionEvents = MagicMock()
     sys.modules['deepgram.clients.live.v1'].LiveOptions = MagicMock
 
+_speaker_embedding = ModuleType('utils.stt.speaker_embedding')
+_speaker_embedding.SPEAKER_MATCH_THRESHOLD = 0.45
+_speaker_embedding.async_extract_embedding_from_bytes = AsyncMock(return_value=None)
+_speaker_embedding.compare_embeddings = MagicMock(return_value=0.0)
+sys.modules.setdefault('utils.stt.speaker_embedding', _speaker_embedding)
+
+_vad = ModuleType('utils.stt.vad')
+_vad._get_ort_session = MagicMock()
+_vad.make_fresh_state = MagicMock(return_value=(None, None))
+_vad.run_vad_window = MagicMock(return_value=0.0)
+_vad.VAD_WINDOW_SAMPLES = 512
+sys.modules.setdefault('utils.stt.vad', _vad)
+
 from utils.stt.streaming import connect_to_deepgram_with_backoff, process_audio_dg  # noqa: E402
 from utils.stt.streaming import deepgram_options, deepgram_cloud_options  # noqa: E402
 from utils.stt.streaming import get_stt_service_for_language, STTService, should_preserve_filler_words  # noqa: E402