Re - feat(pipecat-sdk): add speech-to-speech model support (Gemini Live) (#683)pipecat-update

#### RE-RAISING Pipecat live speech PR

### Added native speech-to-speech model support

### Summary:
  - Speech-to-speech support - Auto-detect audio frames and inject memories to system prompt for native audio models (Gemini Live, etc.)
  - Fix memory bloating - Replace memories each turn using XML tags instead of accumulating
  - Add temporal context - Show recency on search results ([2d ago], [15 Jan])
  - New inject_mode param - auto (default), system, or user

### Docs update
  - Update the docs for native speech-2-speech models

1 files changed, 14 insertions, 13 deletions

diff --git a/packages/pipecat-sdk-python/README.md b/packages/pipecat-sdk-python/README.md
index 5f6e8478..bb8e26e1 100644
--- a/packages/pipecat-sdk-python/README.md
+++ b/packages/pipecat-sdk-python/README.md
@@ -107,11 +107,9 @@ from fastapi import FastAPI, WebSocket
 from pipecat.pipeline.pipeline import Pipeline
 from pipecat.pipeline.task import PipelineTask
 from pipecat.pipeline.runner import PipelineRunner
-from pipecat.services.openai import (
-    OpenAILLMService,
-    OpenAIUserContextAggregator,
-)
-from pipecat.transports.network.fastapi_websocket import (
+from pipecat.processors.aggregators.openai_llm_context import OpenAILLMContext
+from pipecat.services.google.gemini_live.llm import GeminiLiveLLMService
+from pipecat.transports.websocket.fastapi import (
     FastAPIWebsocketTransport,
     FastAPIWebsocketParams,
 )
@@ -125,10 +123,17 @@ async def websocket_endpoint(websocket: WebSocket):
 
     transport = FastAPIWebsocketTransport(
         websocket=websocket,
-        params=FastAPIWebsocketParams(audio_out_enabled=True),
+        params=FastAPIWebsocketParams(audio_in_enabled=True, audio_out_enabled=True),
+    )
+
+    # Gemini Live for speech-to-speech
+    llm = GeminiLiveLLMService(
+        api_key=os.getenv("GEMINI_API_KEY"),
+        model="models/gemini-2.5-flash-native-audio-preview-12-2025",
     )
 
-    user_context = OpenAIUserContextAggregator()
+    context = OpenAILLMContext([{"role": "system", "content": "You are a helpful assistant."}])
+    context_aggregator = llm.create_context_aggregator(context)
 
     # Supermemory memory service
     memory = SupermemoryPipecatService(
@@ -136,17 +141,13 @@ async def websocket_endpoint(websocket: WebSocket):
         session_id="session-123",
     )
 
-    llm = OpenAILLMService(
-        api_key=os.getenv("OPENAI_API_KEY"),
-        model="gpt-4",
-    )
-
     pipeline = Pipeline([
         transport.input(),
-        user_context,
+        context_aggregator.user(),
         memory,
         llm,
         transport.output(),
+        context_aggregator.assistant(),
     ])
 
     runner = PipelineRunner()