Revert "Revert "only append audio if speech has started""

This reverts commit 992ca8f.

openduck-py/openduck_py/response_agent.py

@@ -243,6 +243,7 @@ def __init__(
         self.response_task: Optional[asyncio.Task] = None
         self.interrupt_event = asyncio.Event()
         self.system_prompt = system_prompt
+        self.speech_has_started = False
 
         if context is None:
             context = {}
@@ -308,7 +309,6 @@ async def receive_audio(self, message: bytes):
             audio_16k_np = audio_16k_np.astype(np.float32)
 
         # TODO: Only append to audio_data if start of speech has been detected.
-        self.audio_data.append(audio_16k_np)
         if self.record:
             if self._time_of_last_record is None:
                 self._time_of_last_record = time()
@@ -326,29 +326,32 @@ async def receive_audio(self, message: bytes):
             vad_result = self.vad(audio_16k_chunk)
             if vad_result:
                 async with SessionAsync() as db:
-
                     transcription = ""
                     if "start" in vad_result or "end" in vad_result:
                         self.time_of_last_activity = time()
-                        audio_data = np.concatenate(self.audio_data)
-
                         if "start" in vad_result:
+                            self.speech_has_started = True
                             print("Detected start of speech", flush=True)
-                            await log_event(
-                                db,
-                                self.session_id,
-                                "detected_start_of_speech",
-                                audio=audio_data,
-                            )
+                            # await log_event(
+                            #     db,
+                            #     self.session_id,
+                            #     "detected_start_of_speech",
+                            #     audio=audio_data,
+                            # )
                         else:
+                            self.speech_has_started = False
                             print("Detected end of speech", flush=True)
-                            await log_event(
-                                db,
-                                self.session_id,
-                                "detected_end_of_speech",
-                                audio=audio_data,
-                            )
+                            # await log_event(
+                            #     db,
+                            #     self.session_id,
+                            #     "detected_end_of_speech",
+                            #     audio=audio_data,
+                            # )
+
+                        if self.speech_has_started:
+                            self.audio_data.append(audio_16k_np)
 
+                        audio_data = np.concatenate(self.audio_data)
                         transcription = await _transcribe(audio_data)
                         if not transcription:
                             continue