MoviePy adds sound artifact to the end of the audio file after reading it

[YaroslavSchubert]

Expected Behavior

The resulting audio file will be just a copy of the original one.

Actual Behavior

The resulting audio file has a small artifact added to its end.

Steps to Reproduce the Problem

Here is a code to reproduce the problem:

from moviepy.audio.io.AudioFileClip import AudioFileClip

audio_path = "data/yt_short_sample.mp3"
audio = AudioFileClip(audio_path)
audio.write_audiofile("data/artifacts/test.mp3")

Example audio file:
https://soundcloud.com/yaroslav-schubert/sample-audio-to-reproduce-moviepy-issue/s-IZjaAlFxn3L

Here are the screenshots from Audacity:

End of the original audio file

End of the resulting audio file

Specifications

• Python Version: 3.10.6
• MoviePy Version: 1.0.3
• Platform Name: Linux/Ubuntu
• Platform Version: 18.04.6

[thertzelle]

+1 Thanks for posting this. I thought I was going crazy, or I implemented something odd, but every so often, maybe by the 3rd of 4th clip, its like the first microsecond of the audio is played back before the next file starts. It happens no matter what encoding I use for the audio.

[YaroslavSchubert]

I ended up adding audio to my clips by specifying an audio file path to the audio argument of the write_videofile function:

video_clip.write_videofile(out_file_path, fps=moviepy_fps, audio=audio_path)

[thertzelle]

So you have multiple clips? But then you have an audio_path? I don't get it, the clips have associated audios, then what is the audio_path for?

How do you keep the audio for each clip synced with the clip?

[YaroslavSchubert]

I have a single CompositeVideoClip that combines ImageClip, ColorClip, TextClip , VideoClip. I also have a single audio file that I add to the resulting CompositeVideoClip.

[jake-nz]

This is caused by ffmepg incorrectly reporting the duration of the audio.
This causes iter_chunks in AudioClip to try to read frames outside the length of the file. In FFMPEG_AudioReader get_frame reads the end of the file again, resulting in a glitch.
For me, the fix was to set the env var FFMPEG_BINARY=/usr/bin/ffmpeg to use ffmpeg 6.0 from my system instead of ffmpeg 4.2.2 from inside imageio_ffmpeg

[jake-nz]

Same issue:
#1854
#1005
#1717
#1710
#1936

[lightwastak3n]

@jake-nz
So like this https://colab.research.google.com/drive/1ZKkeWAKJ6GUQ4NuAjID99nbyAiIjE6ix?usp=sharing
It didn't fix it for me. Cutting a bit of the audio at the end and then combining it fixes it for me. But some of my files have speech until the end so it's not always doable.

I thought it was maybe some kind of padding to get the correct fps but I don't see any pattern in length of the audios that produce artifact and the ones that don't.

Update:
I found ~20 of these audio files that produce artifact and cutting 0.05 from their end fixes it for me. (I'm also extending image looping since audio has speech until the very end so this flows nicer)

audio = AudioFileClip(audio_path)
audio = audio.subclip(0, audio.duration - 0.05)

# Combine with image and add pause at the end (optional)
image_clip = ImageClip(image_path, duration=audio.duration + 0.2)
video_clip = image_clip.set_audio(audio)

[gchila]

why is this closed while the problem is not solved ?