Benchmark: ONNX Runtime vs HF Transformers vs GGUF for Parakeet TDT 0.6B on CPU-only hardware [D]

In three linesCPU inference benchmark for Parakeet TDT 0.6B on 2 x86-64 vCPUs (7.7GB RAM). ONNX Runtime FP32 achieves RTF 0.328 (37% faster than HF Transformers bfloat16 at 0.519) but peaks at 2.7GB memory. GGUF Q6_K reduces to 928MB but doubles RTF to 0.708. Methodological note: espeak-ng inflates WER to 20.9% vs gTTS 4.65%.

Summary generated by Claude — human-verified