feat: add useful experiment logs to various agents

ref: N25B-401

src/control_backend/agents/perception/transcription_agent/transcription_agent.py

@@ -1,4 +1,5 @@
 import asyncio
+import logging
 
 import numpy as np
 import zmq
@@ -10,6 +11,8 @@ from control_backend.core.config import settings
 
 from .speech_recognizer import SpeechRecognizer
 
+experiment_logger = logging.getLogger(settings.logging_settings.experiment_logger_name)
+
 
 class TranscriptionAgent(BaseAgent):
     """
@@ -25,6 +28,8 @@ class TranscriptionAgent(BaseAgent):
     :ivar audio_in_socket: The ZMQ SUB socket instance.
     :ivar speech_recognizer: The speech recognition engine instance.
     :ivar _concurrency: Semaphore to limit concurrent transcriptions.
+    :ivar _current_speech_reference: The reference of the current user utterance, for synchronising
+        experiment logs.
     """
 
     def __init__(self, audio_in_address: str):
@@ -39,6 +44,7 @@ class TranscriptionAgent(BaseAgent):
         self.audio_in_socket: azmq.Socket | None = None
         self.speech_recognizer = None
         self._concurrency = None
+        self._current_speech_reference: str | None = None
 
     async def setup(self):
         """
@@ -63,6 +69,10 @@ class TranscriptionAgent(BaseAgent):
 
         self.logger.info("Finished setting up %s", self.name)
 
+    async def handle_message(self, msg: InternalMessage):
+        if msg.thread == "voice_activity":
+            self._current_speech_reference = msg.body
+
     async def stop(self):
         """
         Stop the agent and close sockets.
@@ -96,24 +106,25 @@ class TranscriptionAgent(BaseAgent):
 
     async def _share_transcription(self, transcription: str):
         """
-        Share a transcription to the other agents that depend on it.
+        Share a transcription to the other agents that depend on it, and to experiment logs.
 
         Currently sends to:
         - :attr:`settings.agent_settings.text_belief_extractor_name`
+        - The UI via the experiment logger
 
         :param transcription: The transcribed text.
         """
-        receiver_names = [
-            settings.agent_settings.text_belief_extractor_name,
-        ]
+        experiment_logger.chat(
+            transcription,
+            extra={"role": "user", "reference": self._current_speech_reference, "partial": False},
+        )
 
-        for receiver_name in receiver_names:
-            message = InternalMessage(
-                to=receiver_name,
-                sender=self.name,
-                body=transcription,
-            )
-            await self.send(message)
+        message = InternalMessage(
+            to=settings.agent_settings.text_belief_extractor_name,
+            sender=self.name,
+            body=transcription,
+        )
+        await self.send(message)
 
     async def _transcribing_loop(self) -> None:
         """
@@ -129,10 +140,9 @@ class TranscriptionAgent(BaseAgent):
                 audio = np.frombuffer(audio_data, dtype=np.float32)
                 speech = await self._transcribe(audio)
                 if not speech:
-                    self.logger.info("Nothing transcribed.")
+                    self.logger.debug("Nothing transcribed.")
                     continue
 
-                self.logger.info("Transcribed speech: %s", speech)
                 await self._share_transcription(speech)
             except Exception as e:
                 self.logger.error(f"Error in transcription loop: {e}")