In line 2876 of whiper.cpp, this code will skip some short voice segments incorrectly when speed_up is true. Should the 100 samples here be changed to a smaller value when speed_up is true, such as 50?
// if length of spectrogram is less than 1s (100 samples), then return
// basically don't process anything that is less than 1s
// see issue #39: https://github.com/ggerganov/whisper.cpp/issues/39
if (seek_end < 100 + seek_start) {
return 0;
}