Merge pull request #3212 from pipecat-ai/pk/nova-2-sonic

Nova 2 Sonic support

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

@@ -5,7 +5,9 @@
 #
 
 
+import asyncio
 import os
+import random
 from datetime import datetime
 
 from dotenv import load_dotenv
@@ -33,11 +35,21 @@ load_dotenv(override=True)
 
 
 async def fetch_weather_from_api(params: FunctionCallParams):
-    temperature = 75 if params.arguments["format"] == "fahrenheit" else 24
+    temperature = (
+        random.randint(60, 85)
+        if params.arguments["format"] == "fahrenheit"
+        else random.randint(15, 30)
+    )
+    # Simulate a long network delay.
+    # You can continue chatting while waiting for this to complete.
+    # With Nova 2 Sonic (the default model), the assistant will respond
+    # appropriately once the function call is complete.
+    await asyncio.sleep(5)
     await params.result_callback(
         {
             "conditions": "nice",
             "temperature": temperature,
+            "location": params.arguments["location"],
             "format": params.arguments["format"],
             "timestamp": datetime.now().strftime("%Y%m%d_%H%M%S"),
         }
@@ -91,23 +103,31 @@ 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. "
-        f"{AWSNovaSonicLLMService.AWAIT_TRIGGER_ASSISTANT_RESPONSE_INSTRUCTION}"
+        "two or three sentences for chatty scenarios."
+        # 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
     llm = AWSNovaSonicLLMService(
         secret_access_key=os.getenv("AWS_SECRET_ACCESS_KEY"),
         access_key_id=os.getenv("AWS_ACCESS_KEY_ID"),
-        region=os.getenv("AWS_REGION"),  # as of 2025-05-06, us-east-1 is the only supported region
+        # as of 2025-12-09, these are the supported regions:
+        # - Nova 2 Sonic (the default model):
+        #   - us-east-1
+        #   - us-west-2
+        #   - ap-northeast-1
+        # - Nova Sonic (the older model):
+        #   - us-east-1
+        #   - ap-northeast-1
+        region=os.getenv("AWS_REGION"),
         session_token=os.getenv("AWS_SESSION_TOKEN"),
-        voice_id="tiffany",  # matthew, tiffany, amy
+        voice_id="tiffany",
         # you could choose to pass instruction here rather than via context
         # system_instruction=system_instruction
         # you could choose to pass tools here rather than via context
@@ -117,7 +137,9 @@ async def run_bot(transport: BaseTransport, runner_args: RunnerArguments):
     # Register function for function calls
     # you can either register a single function for all function calls, or specific functions
     # llm.register_function(None, fetch_weather_from_api)
-    llm.register_function("get_current_weather", fetch_weather_from_api)
+    llm.register_function(
+        "get_current_weather", fetch_weather_from_api, cancel_on_interruption=False
+    )
 
     # Set up context and context management.
     context = LLMContext(
@@ -159,10 +181,10 @@ 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
-        # Nova Sonic. Note that this trigger requires a special corresponding bit of text in the
-        # system instruction. In the future, simply queueing the context frame should be sufficient.
-        await llm.trigger_assistant_response()
+        # HACK: if using the older Nova Sonic (pre-2) model, you need this special way of
+        # triggering the first assistant response. Note that this trigger requires a special
+        # corresponding bit of text in the system instruction.
+        # await llm.trigger_assistant_response()
 
     # Handle client disconnection events
     @transport.event_handler("on_client_disconnected")