cut.load_audio() fail bug.

[kobenaxie]

audio.txt
Using this to convert audio.txt to audio.bin, as I can't upload a binary file.

import base64

with open('audio.txt', 'r') as f:
    encoded_data = f.read()

binary_data = base64.b64decode(encoded_data)

with open('audio.bin', 'wb') as f:
    f.write(binary_data)

After converting a recording to cut, cut.load_audio() failed with lhotse==1.30.3.

from lhotse import Recording
from lhotse.audio.backend import audio_backend, TorchaudioDefaultBackend

with open("audio.bin", "rb") as f:
    audio_bytes = f.read()
 
def success():
    with audio_backend(TorchaudioDefaultBackend):
        recording = Recording.from_bytes(audio_bytes, "1")
    cut = recording.to_cut()
    cut.load_audio()

def fail():
    recording = Recording.from_bytes(audio_bytes, "1")
    cut = recording.to_cut()
    cut.load_audio()


if __name__ == "__main__":
    success()
    fail()

Error msg:

Traceback (most recent call last):
  File "/conda_envs/torch/lib/python3.10/site-packages/lhotse/utils.py", line 848, in wrapper
    return fn(*args, **kwargs)
  File "/conda_envs/torch/lib/python3.10/site-packages/lhotse/audio/recording.py", line 485, in load_audio
    audio = assert_and_maybe_fix_num_samples(
  File "/conda_envs/torch/lib/python3.10/site-packages/lhotse/audio/recording.py", line 969, in assert_and_maybe_fix_num_samples
    raise AudioLoadingError(
lhotse.audio.utils.AudioLoadingError: The number of declared samples in the recording diverged from the one obtained when loading audio (offset=0.0, duration=3.24715625). This could be internal Lhotse's error or a faulty transform implementation. Please report this issue in Lhotse and show the following: diff=1381, audio.shape=(1, 102528), recording=Recording(id='1', sources=[AudioSource(type='memory', channels=[0], source='<binary-data>')], sampling_rate=32000, num_samples=103909, duration=3.24715625, channel_ids=[0], transforms=None)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/conda_envs/torch/lib/python3.10/site-packages/lhotse/utils.py", line 848, in wrapper
    return fn(*args, **kwargs)
  File "/conda_envs/torch/lib/python3.10/site-packages/lhotse/cut/mono.py", line 77, in load_audio
    return self.recording.load_audio(
  File "/conda_envs/torch/lib/python3.10/site-packages/lhotse/utils.py", line 850, in wrapper
    raise type(e)(
lhotse.audio.utils.AudioLoadingError: The number of declared samples in the recording diverged from the one obtained when loading audio (offset=0.0, duration=3.24715625). This could be internal Lhotse's error or a faulty transform implementation. Please report this issue in Lhotse and show the following: diff=1381, audio.shape=(1, 102528), recording=Recording(id='1', sources=[AudioSource(type='memory', channels=[0], source='<binary-data>')], sampling_rate=32000, num_samples=103909, duration=3.24715625, channel_ids=[0], transforms=None)
[extra info] When calling: Recording.load_audio(args=(Recording(id='1', sources=[AudioSource(type='memory', channels=[0], source='<binary-data>')], sampling_rate=32000, num_samples=103909, duration=3.24715625, channel_ids=[0], transforms=None),) kwargs={'channels': 0, 'offset': 0.0, 'duration': 3.24715625})

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "./bug.py", line 23, in <module>
    fail()
  File "./bug.py", line 17, in fail
    cut.load_audio()
  File "/conda_envs/torch/lib/python3.10/site-packages/lhotse/utils.py", line 850, in wrapper
    raise type(e)(
lhotse.audio.utils.AudioLoadingError: The number of declared samples in the recording diverged from the one obtained when loading audio (offset=0.0, duration=3.24715625). This could be internal Lhotse's error or a faulty transform implementation. Please report this issue in Lhotse and show the following: diff=1381, audio.shape=(1, 102528), recording=Recording(id='1', sources=[AudioSource(type='memory', channels=[0], source='<binary-data>')], sampling_rate=32000, num_samples=103909, duration=3.24715625, channel_ids=[0], transforms=None)
[extra info] When calling: Recording.load_audio(args=(Recording(id='1', sources=[AudioSource(type='memory', channels=[0], source='<binary-data>')], sampling_rate=32000, num_samples=103909, duration=3.24715625, channel_ids=[0], transforms=None),) kwargs={'channels': 0, 'offset': 0.0, 'duration': 3.24715625})
[extra info] When calling: MonoCut.load_audio(args=(MonoCut(id='1', start=0.0, duration=3.24715625, channel=0, supervisions=[], features=None, recording=Recording(id='1', sources=[AudioSource(type='memory', channels=[0], source='<binary-data>')], sampling_rate=32000, num_samples=103909, duration=3.24715625, channel_ids=[0], transforms=None), custom=None),) kwargs={})