encode time and decode time is cost 4x-5x times the same environment and the same model in my demo

[wuhongsheng]

./main -m ./models/ggml-medium.bin -f ./samples/jfk.wav

whisper_print_timings: load time = 3692.32 ms whisper_print_timings: mel time = 48.12 ms whisper_print_timings: sample time = 13.55 ms whisper_print_timings: encode time = 13432.69 ms / 559.70 ms per layer whisper_print_timings: decode time = 1076.62 ms / 44.86 ms per layer whisper_print_timings: total time = 18268.29 ms

in my demo whisper_print_timings: load time = 3322.22 ms whisper_print_timings: mel time = 59.26 ms whisper_print_timings: sample time = 14.02 ms whisper_print_timings: encode time = 64995.89 ms / 2708.16 ms per layer whisper_print_timings: decode time = 3828.06 ms / 159.50 ms per layer whisper_print_timings: total time = 72222.48 ms