Agent 1.0 voice agent does not work properly with some of ARM64 CPUs

[prasanna4742]

I'm running voice agent in EKS with node as - c8g.large, ARM64 based AWS's graviton instance. When the agent starts, it never receives human speech and it emits below warnings:

` 2025-05-16 08:20:57 - livekit.agents - WARNING - [RM_po4rSeXLQdTy] Running <Task pending name='job_shutdown_callback' coro=<JobContext.add_shutdown_callback..wrapper() running at /usr/local/lib/python3.12/site-packages/livekit/agents/job.py:226> wait_for=<Task pending name='Task-311' coro=<RunnableBindingBase.ainvoke() running at /usr/local/lib/python3.12/site-packages/langchain_core/runnables/base.py:5364> cb=[Task.task_wakeup()]> cb=[gather.._done_callback() at /usr/local/lib/python3.12/asyncio/tasks.py:767]> took too long: 19.66 seconds 2025-05-16 08:20:57 - livekit.plugins.silero - WARNING - [RM_po4rSeXLQdTy] inference is slower than realtime 2025-05-16 08:20:57 - livekit.plugins.silero - WARNING - [RM_po4rSeXLQdTy] inference is slower than realtime 2025-05-16 08:20:58 - livekit.plugins.silero - WARNING - [RM_po4rSeXLQdTy] inference is slower than realtime 2025-05-16 08:20:58 - livekit.plugins.silero - WARNING - [RM_po4rSeXLQdTy] inference is slower than realtime 2025-05-16 08:20:58 - livekit.plugins.silero - WARNING - [RM_po4rSeXLQdTy] inference is slower than realtime

` Exact same code works fine when the node is AMD64 based node like a C5A machine.

This suggests that either VAD or agent code does not work properly on ARM64 based processors. But strange part is same code works fine on my local Mac M3 Pro which is ARM64 based.

So probably some issue related to AWS graviton instances?