🧹 Decouple HTTP exceptions from VAD utility

backend/routers/speech_profile.py

@@ -21,7 +21,7 @@
 )
 from utils.multipart import MultipartMaxPartSizeRoute, SPEECH_PROFILE_MAX_PART_SIZE, max_part_size
 from utils.stt.speaker_embedding import extract_embedding
-from utils.stt.vad import apply_vad_for_speech_profile
+from utils.stt.vad import apply_vad_for_speech_profile, VADEmptyError
 import logging
 
 logger = logging.getLogger(__name__)
@@ -81,7 +81,10 @@ def upload_profile(file: UploadFile, uid: str = Depends(auth.get_current_user_ui
     if aseg.duration_seconds < 5 or aseg.duration_seconds > 120:
         raise HTTPException(status_code=400, detail="Audio duration is invalid (must be 5-120 seconds)")
 
-    apply_vad_for_speech_profile(file_path)
+    try:
+        apply_vad_for_speech_profile(file_path)
+    except VADEmptyError:
+        raise HTTPException(status_code=400, detail="Audio is empty")
 
     # Write-ahead: Cache exact duration after VAD processing (use av for fast header-only read)
     with av.open(file_path) as container:

backend/scripts/stt/j_apply_vad_to_speech_profiles.py

@@ -7,7 +7,7 @@
 from dotenv import load_dotenv
 from pydub import AudioSegment
 
-from utils.stt.vad import apply_vad_for_speech_profile
+from utils.stt.vad import apply_vad_for_speech_profile, VADEmptyError
 
 load_dotenv('../../.env')
 os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = '../../' + os.getenv('GOOGLE_APPLICATION_CREDENTIALS', '')
@@ -25,7 +25,13 @@ def single(uid: str) -> None:
         file_path = get_profile_audio_if_exists(uid)
         if not file_path:
             return
-        apply_vad_for_speech_profile(file_path)
+
+        try:
+            apply_vad_for_speech_profile(file_path)
+        except VADEmptyError:
+            print('VAD empty for', uid)
+            return
+
         aseg = cast(
             Any, AudioSegment.from_wav(file_path)
         )  # pyright: ignore[reportUnknownMemberType]  # pydub has no type stubs

backend/tests/unit/test_speech_profile_wav_decode.py

@@ -12,7 +12,6 @@
 
 import importlib.abc
 import importlib.machinery
-import importlib.util
 import io
 import os
 import sys
@@ -29,9 +28,7 @@
     "database",
     "utils.other.storage",
     "utils.stt.speaker_embedding",
-    "utils.stt.vad",
     "av",
-    "pydub",
     "firebase_admin",
     "google",
     "pinecone",
@@ -75,6 +72,8 @@ def exec_module(self, module):
 sys.meta_path.insert(0, _f)
 try:
     from routers import speech_profile as mod
+    from scripts.stt import j_apply_vad_to_speech_profiles as batch_mod
+    from utils.stt import vad as vad_mod
 finally:
     sys.meta_path.remove(_f)
     for _n in list(sys.modules):
@@ -132,3 +131,35 @@ def test_invalid_wav_does_not_run_vad_or_upload(self):
         assert exc_info.value.status_code == 400
         mock_vad.assert_not_called()
         mock_upload.assert_not_called()
+
+    def test_empty_vad_returns_400_not_500(self):
+        fake_file = _fake_upload_file(b'silence')
+
+        with patch.object(mod, "os") as mock_os, patch("builtins.open", MagicMock()), patch.object(
+            mod, "AudioSegment"
+        ) as mock_aseg, patch.object(vad_mod, "vad_is_empty", return_value=[]) as mock_vad, patch.object(
+            mod, "upload_profile_audio"
+        ) as mock_upload:
+            mock_os.makedirs.return_value = None
+            mock_aseg.from_wav.return_value = MagicMock(frame_rate=16000, duration_seconds=5)
+
+            with pytest.raises(HTTPException) as exc_info:
+                mod.upload_profile(fake_file, uid="test-uid")
+
+        assert exc_info.value.status_code == 400
+        assert exc_info.value.detail == "Audio is empty"
+        mock_vad.assert_called_once_with("_temp/test-uid/speech_profile.wav", return_segments=True)
+        mock_upload.assert_not_called()
+
+    def test_batch_skips_empty_vad_without_uploading(self):
+        with patch.object(batch_mod.os, "makedirs"), patch.object(
+            batch_mod, "get_users_uid", return_value=["test-uid"]
+        ), patch.object(batch_mod, "get_profile_audio_if_exists", return_value="/tmp/profile.wav"), patch.object(
+            batch_mod, "upload_profile_audio"
+        ) as mock_upload, patch.object(
+            vad_mod, "vad_is_empty", return_value=[]
+        ) as mock_vad:
+            batch_mod.execute()
+
+        mock_vad.assert_called_once_with("/tmp/profile.wav", return_segments=True)
+        mock_upload.assert_not_called()

backend/tests/unit/test_vad_onnx.py

@@ -23,6 +23,7 @@
 from utils.stt.vad import (
     VADAudioDecodeError,
     VADProcessingError,
+    VADEmptyError,
     vad_is_empty,
     _run_file_vad,
     _get_ort_session,
@@ -36,6 +37,13 @@
     _STATE_SHAPE,
 )
 
+
+def test_apply_vad_for_speech_profile_raises_for_zero_segments():
+    with patch.object(vad, 'vad_is_empty', return_value=[]):
+        with pytest.raises(VADEmptyError, match='Audio is empty'):
+            vad.apply_vad_for_speech_profile('/fake/path.wav')
+
+
 # ---------------------------------------------------------------------------
 # Helpers
 # ---------------------------------------------------------------------------

backend/utils/stt/vad.py

@@ -8,7 +8,6 @@
 import numpy as np
 import onnxruntime as ort  # onnxruntime is untyped
 import requests
-from fastapi import HTTPException
 from pydub import AudioSegment  # pydub is untyped
 
 from database import redis_db
@@ -34,6 +33,10 @@ class VADProcessingError(RuntimeError):
     """The local VAD could not make a trustworthy speech decision."""
 
 
+class VADEmptyError(ValueError):
+    """The VAD found no speech segments in the audio."""
+
+
 def _positive_timeout_seconds(env_name: str, default: float, maximum: float) -> float:
     raw_value = os.getenv(env_name)
     if raw_value is None:
@@ -395,8 +398,8 @@ async def async_vad_is_empty(
 def apply_vad_for_speech_profile(file_path: str) -> None:
     logger.info(f'apply_vad_for_speech_profile {file_path}')
     voice_segments = vad_is_empty(file_path, return_segments=True)
-    if len(voice_segments) == 0:  # TODO: front error on post-processing, audio sent is bad.
-        raise HTTPException(status_code=400, detail="Audio is empty")
+    if len(voice_segments) == 0:
+        raise VADEmptyError("Audio is empty")
     joined_segments: List[Dict[str, Any]] = []
     for i, segment in enumerate(voice_segments):
         if joined_segments and (segment['start'] - joined_segments[-1]['end']) < 1: