feat(elevenlabs): add eleven_v3 model support with HTTP streaming fallback

examples/eleven_v3_example.py

@@ -0,0 +1,45 @@
+"""
+Example demonstrating ElevenLabs eleven_v3 TTS model usage with LiveKit Agents.
+
+The eleven_v3 model doesn't support WebSocket streaming, so the plugin automatically
+uses HTTP streaming with chunked transfer encoding instead.
+
+To compare latency between models:
+- Set USE_V3 = True  → Uses eleven_v3 with HTTP streaming (new)
+- Set USE_V3 = False → Uses eleven_turbo_v2_5 with WebSocket (existing)
+
+Look for the "tts_ttfb" metric in the console to compare TTS latency.
+"""
+
+from livekit.agents import AgentServer, JobContext, cli
+from livekit.agents.voice import Agent, AgentSession
+from livekit.plugins import elevenlabs, openai, silero
+
+server = AgentServer()
+
+
+@server.rtc_session()
+async def entrypoint(ctx: JobContext):
+    # Toggle between models to compare latency:
+    # - eleven_v3: Uses HTTP streaming (our new implementation)
+    # - eleven_turbo_v2_5: Uses WebSocket streaming (existing)
+
+    USE_V3 = True  # Set to False to test eleven_turbo_v2_5
+
+    agent = Agent(
+        instructions="You are a helpful voice assistant. Keep responses very short - 1 sentence max.",
+        stt=openai.STT(),  # OpenAI Whisper
+        llm=openai.LLM(model="gpt-4o"),  # Faster than gpt-4o-mini for streaming
+        tts=elevenlabs.TTS(
+            model="eleven_v3" if USE_V3 else "eleven_turbo_v2_5",
+            voice_id="EXAVITQu4vr4xnSDxMaL",
+        ),
+        vad=silero.VAD.load(),
+    )
+
+    session = AgentSession()
+    await session.start(agent=agent, room=ctx.room)
+
+
+if __name__ == "__main__":
+    cli.run_app(server)

livekit-plugins/livekit-plugins-elevenlabs/README.md

@@ -13,3 +13,13 @@ pip install livekit-plugins-elevenlabs
 ## Pre-requisites
 
 You'll need an API key from ElevenLabs. It can be set as an environment variable: `ELEVEN_API_KEY`
+
+## Supported Models
+
+All ElevenLabs TTS models are supported, including:
+- `eleven_v3` - Most expressive model with emotion and delivery control
+- `eleven_turbo_v2_5` - Fast, high-quality multilingual model (default)
+- `eleven_flash_v2_5`, `eleven_flash_v2` - Ultra-fast models
+- And more...
+
+**Note:** The `eleven_v3` model uses HTTP streaming instead of WebSocket for compatibility, as it doesn't support WebSocket connections. The plugin automatically handles this difference.