Lossless cutting

[AhmedThahir]

I have taken the Ruby code from #165 and converted into Python.

I had to make a few changes due to changes in the auto-editor API. Would love to get someone's input if this is correct for the latest version.

This is the code I'm trying, but I'm not able to get it work. There are no errors, just that the output is not what I'm expecting; there are black frames.

Maybe @genegoykhman can you help with this?

import re
import sys
import json
import os

def cleanup_pre():
    for file in os.listdir():
        for unwanted in ["_segment", "_stripped"]:
            if unwanted in file.lower():
                os.remove(file)

def cleanup_post():
    for file in os.listdir():
        for unwanted in ["_segment"]:
            if unwanted in file.lower():
                os.remove(file)

def fps(input_file):
    info = os.popen(f'ffmpeg -i "{input_file}" 2>&1').read()
    match = re.search(r'\s([\d\.]*)\sfps', info)
    if match:
        return float(match.group(1))
    return 0.0

def process_json(input_file, fps, json):
    extension = os.path.splitext(input_file)[1]
    with open('_segments.txt', 'w') as f:
        sounded_chunks = json["v"][0]
        for segment_number, sounded_chunk in enumerate(sounded_chunks):
            offset_time = -sounded_chunk["offset"] / fps
            start_time = (sounded_chunk["start"]) / fps
            duration = sounded_chunk["dur"] / fps
            end_time = start_time + duration
            print(sounded_chunk["offset"], sounded_chunk["start"], sounded_chunk["dur"])
            cmd = f"""
            ffmpeg -ss {start_time} -itsoffset {offset_time}  -i "{input_file}" -t {duration} -c copy _segment{segment_number}{extension} -hide_banner -loglevel error
            """
            # print(cmd)
            os.system(cmd)
            f.write(f'file _segment{segment_number}{extension}\n')

def combine_segments(input_file):
    os.system(f"""
    ffmpeg -f concat -safe 0 -i _segments.txt -c copy "{os.path.splitext(input_file)[0]}_STRIPPED{os.path.splitext(input_file)[1]}" -hide_banner -loglevel error
    """)

def process_file(input_file):
    fps_val = fps(input_file)
    if fps_val <= 0.0:
        print(f'Unable to determine FPS of {input_file}')
        return
    os.system(f'auto-editor "{input_file}" --export json')
    json_file = os.path.splitext(input_file)[0] + '_ALTERED.json'
    with open(json_file) as f:
        json_data = json.load(f)
    process_json(input_file, fps_val, json_data)
    os.remove(json_file)
    combine_segments(input_file)

def iterate_files():
    for input_file in sys.argv[1:]:
        process_file(input_file)

def main():
  cleanup_pre()
  iterate_files()
#   cleanup_post()
if __name__ == "__main__":
  main()

[AhmedThahir]

I tried re-encoding the intermediate files (without -c copy)

ffmpeg -ss {start_time} -itsoffset {offset_time}  -i "{input_file}" -t {duration} _segment{segment_number}{extension} -hide_banner -loglevel error

and observed that the intermediate files generated are not what is expected.

I believe this is what is causing the issue

[genegoykhman]

Hi Ahmed,

I can't help too much with the Python code but I'm using a different set of params for ffmpeg. Try converting this to Python and trying it to see if it helps. Note that I'm using the -to param to specify an end-time instead of -itsoffset to specify an offset.

cmd = "ffmpeg -i \"#{input_file}\" -ss #{start_time} -to #{end_time} -c copy _segment#{segment_number}#{extension}""

Here's the full process_json func I'm using right now.

def process_json(input_file, fps, json)
  extension = File.extname(input_file)
  segment_number = 1
  File.open("_segments.txt", "w") do |f|
    json["chunks"].each do |chunk|    
      if chunk[2] == 1
        start_time = chunk[0] / fps
        end_time = chunk[1] / fps
        cmd = "ffmpeg -i \"#{input_file}\" -ss #{start_time} -to #{end_time} -c copy _segment#{segment_number}#{extension}"
        `#{cmd}`
        f.write("file _segment#{segment_number}#{extension}\n")
        segment_number += 1
      end
    end
  end
end

[AhmedThahir]

Hi Gene, I had used your code for reference, but it didn't work.

I observed that you are using chuck[0] and chunk[1], but there are 3 relevant parameters in the export json file, right? Start, Duration, Offset

[AhmedThahir]

Isn't auto-editor's json API 1 deprecated?

[WyattBlue]

No, v1 is stable, not deprecated.

[AhmedThahir]

I tried it, but when I check with ffprobe:

[STREAM]
start_time=0.157000
duration=31.073133
[/STREAM]
[STREAM]
start_time=0.000000
duration=20.275034
[/STREAM]

[AhmedThahir]

I found that this is because: the intermediate clips generated using -c copy do not have equal video and audio length. What should I do for this?

[AhmedThahir]

#400