Optimizing Streaming Speech Recognition on Arm® Edge Devices with ExecuTorch

Each build retained XNNPACK and KleidiAI, and used identical model artifacts. We tested two builds of ExecuTorch 1.3.1, toggling SME2 support on and off. Each configuration had two warm-ups and one measured tracing run. Audio still has to arrive before it can be decoded, so this headroom does not remove the chunk-collection delay.