Audio output bug in LoadVideoDirectory V2

[Jonseed]

When using LoadVideoDirectoryV2 to extract audio, the node fails with this ffmpeg error:

[out#0/f32le @ ...] Output with label 's0:a' does not exist in any defined filter graph
Error opening output files: Invalid argument
!!! Exception during processing !!! 'NoneType' object is not subscriptable

The problem occurs because the node’s load_audio() function chains .audio onto an ffmpeg input stream after applying ss (seek) and/or an atrim filter.
When trimming is applied, the ffmpeg-python graph loses its reference to the labeled audio stream (s0:a), which causes ffmpeg to throw this error and return None.
That None value later triggers a crash in the downstream save_audio node.

---

How to Reproduce

1. Use LoadVideoDirectoryV2 on any MP4 with an audio track.
2. Enable skip_frames or use a short clip (so trimming occurs).
3. Connect the node’s audio output to a Save Audio node.
4. Observe the s0:a error and NoneType traceback.

---

Fix

Remove the .audio chain from the load_audio() function so that the audio stream remains correctly linked to the trimmed input.
Replace:

process = (stream.audio
          .output('pipe:', format='f32le', acodec='pcm_f32le', ac=2, ar=44100)
          .overwrite_output()
          .run_async(pipe_stdout=True, pipe_stderr=True))

with:

process = (
    stream
    .output('pipe:', format='f32le', acodec='pcm_f32le', ac=2, ar=44100, vn=None)
    .overwrite_output()
    .run_async(pipe_stdout=True, pipe_stderr=True)
)

This keeps trimming (ss / atrim) intact, avoids the missing stream label issue, and works across all tested MP4s.

---

Result

After this change:

• Audio extraction works correctly, including trimmed segments.
• No more s0:a or NoneType errors.
• Downstream Save Audio node functions normally.