6th

stores/providers/aaistt.py

@@ -9,6 +9,11 @@ config = aai.TranscriptionConfig(
     language_detection=True,   # auto-detect language
     speaker_labels=True,       # diarization
 )
+_NO_SPEECH_ERRORS = (
+    "no spoken audio",
+    "language_detection cannot be performed",
+    "audio duration is too short",
+)
 
 class AssemblyAISTT:
     def __init__(self):
@@ -21,11 +26,15 @@ class AssemblyAISTT:
         def sync_transcribe():
             transcript = self.client.transcribe(audio_path,config=config)
             if transcript.status == aai.TranscriptStatus.error:
-                raise Exception(f"Transcription failed: {transcript.error}")
+                error_msg = (transcript.error or "").lower()
+
+                # Silence — return empty so the factory raises NoSpeechDetected
+                if any(phrase in error_msg for phrase in _NO_SPEECH_ERRORS):
+                    return "", ""
+
+                # Real error — let the factory wrap it as ProviderUnavailable
+                raise Exception(f"AssemblyAI transcription failed: {transcript.error}")
 
-            text = transcript.text
-            # Language detection is available in transcript.language_code
-            language = transcript.language_code if transcript.language_code else "unknown"
-            return text, language
+            return transcript.text or "", transcript.language_code or "unknown"
 
         return await loop.run_in_executor(None, sync_transcribe)

stores/sttremotes.py

@@ -5,6 +5,12 @@ from .providers.deepgramstt import DeepgramSTT
 from .providers.aaistt import AssemblyAISTT
 from .providers.customstt import CustomSTT
 
+class NoSpeechDetected(Exception):
+    pass
+class ProviderUnavailable(Exception):
+    pass
+
+
 class STTFactory:
     _providers: Dict[str, Type] = {
         "mistral": MistralSTT,
@@ -25,7 +31,21 @@ class STTRemoteManager:
     def __init__(self, default_provider: str = "mistral"):
         self.default_provider = default_provider
 
-    async def transcribe_remote(self, audio_path: str, provider_name: str = None) -> str:
-        provider_name = provider_name or self.default_provider
+    async def transcribe_remote(self, audio_path: str, provider_name: str = None):
+        provider_name = (provider_name or self.default_provider).lower()
         provider = STTFactory.get_provider(provider_name)
-        return await provider.transcribe(audio_path)
+
+        try:
+            text, language = await provider.transcribe(audio_path)
+        except Exception as e:
+            # Real provider error — not silence
+            raise ProviderUnavailable(
+                f"Provider '{provider_name}' failed: {e}"
+            ) from e
+
+        if not text or not text.strip():
+            raise NoSpeechDetected(
+                f"Provider '{provider_name}' returned no speech for this chunk."
+            )
+
+        return text, language