A English file Can't be transcribed correctly with Whisper

[lixikun]

HI , when I test the english model ( small.en or small ) , I get a few text after transcribeing the audio, I check the audio file ,but it is clear. So anyone can help me test this file, this file is come from the gigaspeech , and I convert to short wav file using ffmpeg.
the test code:

import time
import whisper
model=whisper.load_model("small.en").cuda()
time_beg = time.time()
options = {"task":"transcribe","language":"en","fp16":True,'no_speech_threshold':0.1, "condition_on_previous_text": True, "logprob_threshold": -1.00, "without_timestamps":True}
print(options)
#result = model.transcribe("23.wav", options)
result=whisper.transcribe(model, audio="10.wav", verbose=True, **options)
time_end = time.time()
forward_time = time_end - time_beg
print(forward_time)
print(result)

the log:

[00:00.000 --> 00:10.000]  CHAPTER I-III.
1.2601008415222168
{'text': ' CHAPTER I-III.', 'segments': [{'id': 0, 'seek': 0, 'start': 0.0, 'end': 10.0, 'text': ' CHAPTER I-III.', 'tokens': [5870, 29485, 314, 12, 10855, 13], 'temperature': 0.0, 'avg_logprob': -0.652841431753976, 'compression_ratio': 0.6363636363636364, 'no_speech_prob': 0.02467663213610649}], 'language': 'en'}

test file is on the google drive.
10.wav

after I test on the medium model, still get the same results:
the log info with medium model

Detected language 'en' with probability 0.990723
[0.00s -> 10.00s]  CHAPTER XIII.
2.153162956237793

The true result is below, why is it so difference ??? :
chapters one to three of hebrews. this is a library box recording. all library box recordings are in the public domain.for more information or to volunteer, please visit librivox dot org. recording by patty cunningham. hebrews: king james version.chapter one.god who at sundry?