Changes related to Nova 2 Sonic's support for the model speaking first

examples/foundational/40-aws-nova-sonic.py

@@ -91,14 +91,14 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     logger.info(f"Starting bot")
 
     # Specify initial system instruction.
-    # HACK: note that, for now, we need to inject a special bit of text into this instruction to
-    # allow the first assistant response to be programmatically triggered (which happens in the
-    # on_client_connected handler, below)
     system_instruction = (
         "You are a friendly assistant. The user and you will engage in a spoken dialog exchanging "
         "the transcripts of a natural real-time conversation. Keep your responses short, generally "
-        "two or three sentences for chatty scenarios. "
+        "two or three sentences for chatty scenarios."
-        f"{AWSNovaSonicLLMService.AWAIT_TRIGGER_ASSISTANT_RESPONSE_INSTRUCTION}"
+        # HACK: if using the older Nova Sonic (pre-2) model, note that you need to inject a special
+        # bit of text into this instruction to allow the first assistant response to be
+        # programmatically triggered (which happens in the on_client_connected handler)
+        # f"{AWSNovaSonicLLMService.AWAIT_TRIGGER_ASSISTANT_RESPONSE_INSTRUCTION}"
     )
 
     # Create the AWS Nova Sonic LLM service
@@ -167,10 +167,10 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
         logger.info(f"Client connected")
         # Kick off the conversation.
         await task.queue_frames([LLMRunFrame()])
-        # HACK: for now, we need this special way of triggering the first assistant response in AWS
+        # HACK: if using the older Nova Sonic (pre-2) model, you need this special way of
-        # Nova Sonic. Note that this trigger requires a special corresponding bit of text in the
+        # triggering the first assistant response. Note that this trigger requires a special
-        # system instruction. In the future, simply queueing the context frame should be sufficient.
+        # corresponding bit of text in the system instruction.
-        await llm.trigger_assistant_response()
+        # await llm.trigger_assistant_response()
 
     # Handle client disconnection events
     @transport.event_handler("on_client_disconnected")

src/pipecat/services/aws/nova_sonic/llm.py

@@ -618,10 +618,18 @@ class AWSNovaSonicLLMService(LLMService):
         )
         return BedrockRuntimeClient(config=config)
 
+    def _is_first_generation_sonic_model(self) -> bool:
+        # Nova Sonic (the older model) is identified by "amazon.nova-sonic-v1:0"
+        return self._model == "amazon.nova-sonic-v1:0"
+
     def _is_endpointing_sensitivity_supported(self) -> bool:
         # endpointing_sensitivity is only supported with Nova 2 Sonic (and,
         # presumably, future models)
-        return self._model != "amazon.nova-sonic-v1:0"
+        return not self._is_first_generation_sonic_model()
+
+    def _is_assistant_response_trigger_needed(self) -> bool:
+        # Assistant response trigger audio is only needed with the older model
+        return self._is_first_generation_sonic_model()
 
     #
     # LLM communication: input events (pipecat -> LLM)
@@ -1230,7 +1238,8 @@ class AWSNovaSonicLLMService(LLMService):
         )
 
     #
-    # assistant response trigger (HACK)
+    # assistant response trigger
+    # HACK: only needed for the older Nova Sonic (as opposed to Nova 2 Sonic) model
     #
 
     # Class variable
@@ -1244,12 +1253,17 @@ class AWSNovaSonicLLMService(LLMService):
 
         Sends a pre-recorded "ready" audio trigger to prompt the assistant
         to start speaking. This is useful for controlling conversation flow.
-
-        Returns:
-            False if already triggering a response, True otherwise.
         """
+        if not self._is_assistant_response_trigger_needed():
+            logger.warning(
+                f"Assistant response trigger not needed for model '{self._model}'; skipping. "
+                "An LLMRunFrame() should be sufficient to prompt the assistant to respond, "
+                "assuming the context ends in a user message."
+            )
+            return
+
         if self._triggering_assistant_response:
-            return False
+            return
 
         self._triggering_assistant_response = True