`muspy` truncates MIDI files' lengths when converted to numpy arrays

[RishabhMalviya]

I'm trying to convert a MIDI file into a PyTorch Tensor, but muspy seems to be truncating the file.

This is the file's Music representation (after reading it in with test_music = muspy.read_midi('...'):

metadata:
  schema_version: '0.2'
  title: Aeri Aali - Sthaayi 1.1_2
  source_filename: Aeri Aali - Sthaayi 1.1_2.mid
  source_format: midi
resolution: 96
time_signatures:
- time: 0
  numerator: 4
  denominator: 4
- time: 0
  numerator: 4
  denominator: 4
tracks:
- program: 0
  is_drum: false
  notes:
  - time: 0
    pitch: 72
    duration: 24
    velocity: 93
  - time: 21
    pitch: 73
    duration: 33
    velocity: 49
  ...
  - time: 672
    pitch: 68
    duration: 21
    velocity: 96
  - time: 717
    pitch: 68
    duration: 42
    velocity: 119

The last note ends at timestep 759 (0-indexed). So, when I convert this Music object to a piano roll representation using muspy.to_representation(test_music, kind='piano-roll'), I get an array of length 760:

array([[0, 0, 0, ..., 0, 0, 0],
       [0, 0, 0, ..., 0, 0, 0],
       [0, 0, 0, ..., 0, 0, 0],
       ...,
       [0, 0, 0, ..., 0, 0, 0],
       [0, 0, 0, ..., 0, 0, 0],
       [0, 0, 0, ..., 0, 0, 0]], shape=(760, 128), dtype=uint8)

But the original MIDI clip was 768 timesteps long (96 timesteps/beat * 8 beats = 768 timesteps).

How can I get muspy to give me a Tensor that is 768 timesteps long? This is important to my application as the notes need to be placed in the context of the full beat lengths.