0

I have an international conference video which contains two spoken languages, i.e. the video is mixed with sentences of English and Chinese. I would like to remove the Chinese part by command line.

firstly, enerate subtitle files using whisper

whisper myvideo --model large --language en

the subtitle file contains both languages and timing

1
00:00:00,000 --> 00:00:04,220
if you are not concerned and doing the work of the Lord.

2
00:00:09,120 --> 00:00:13,880
如果你不愿意去遵行耶稣基督的话语的话,


3
00:00:14,220 --> 00:00:18,220
就没有必要昼夜去默想神的话。


4
00:00:18,220 --> 00:00:22,200
Take more of me, give me more of you.


....

The question is how to use command line and ffmpeg to remove all the Chinese video parts based on the timing in the subtitle? The video is very long, and the purpose is to use command line to do the task, rather than manually.

step1 ) So I need to identify language of every line of the subtitle:

#!/bin/bash
while IFS= read -r line
do
  echo "text: $line"
  lan= trans -id $line |awk '/^Code/ {print $2}'
  echo "lan: $lan"
done < "$1"

Well this above bash doesn't work properly yet, how to do this?

John Paul Qiang Chen
  • 153
  • What did you try? This is not a command/script writing service. There should be a lot of material regarding how to cut video based on timestamps, which is one of the most basic functionalities. E.g. the one below or some other based on various formats of files. – Destroy666 Feb 12 '24 at 09:01
  • Does this answer your question? Using ffmpeg to cut up video – Destroy666 Feb 12 '24 at 09:01
  • @Destroy666 Thanks a lot, sorry for my narrative, I knew what you suggested. I'm looking for ways to cut video iteratively, the video is large you know. – John Paul Qiang Chen Feb 12 '24 at 09:18
  • You should in that case provide your non-iterative approach and ask how to do it faster, assuming it's slow. But right now, based on what you added, it seems that you're looking for a way to indentify language in CLI, rather. Which should be entirely different question. – Destroy666 Feb 12 '24 at 09:28

1 Answers1

0

I wrote a bash script, barely to remove the Chinese part of the video.

  1. use openai-whisper to generate subtitle file with English and Chinese timing
whisper myvideo -model large-v3 -language Chinese
  1. use bash to extract the desired English timing.
  2. use ffmpeg to extract and merge the English video

A full script of step 2 and 3 is here:

#!/bin/bash
# Date: 20240214
# c275633094@gmail.com
# usage: bash video_rm_chi.sh myvideo.*
# output: *_eng.srt
# output: *_t.txt
# output: *_list.txt
# output: *_list.*
# output: *_eng.*

re_encode=false
clean_final=false


fullfile="$1"
filename=$(basename -- "$fullfile")
extension="${filename##.}"
filename="${filename%.}"


rm "$filename"_t.txt 
rm "$filename"_list*
rm "$filename"_eng*


N=$(wc "$filename".srt |awk '{print $1}')
echo "N= $N"
N4=expr $N / 4
echo "N/4= $N4"
#Nr=expr $N % 4
#echo "N%4= $Nr"


n=1
jump=1
eng=0
tstart=0
tend=0
#while IFS= read -r line
#while read -r line
while(($n<=$N4))
do
  n4=expr $n \* 4 - 1
  echo "n = $n , n4 = $n4"
  if sed "$n4!d" "$filename".srt | ugrep "[\x{4e00}-\x{9fcc}]"; then 
    echo "line expr $n \* 4 - 1 contains Chinese, reject, and jump to next (x4) lyrics line"
    if [ $eng -gt 0 ]; then
      line_eng=$(sed -n "expr $n4 - 5p" "$filename".srt)
      time_eng=($line_eng)
      t2=${time_eng[2]}
      echo "t2 $t2"
    fi
    jump=expr $jump + 1 
    eng=0
  else
    if [ $jump -gt 0 ] && [ $eng -eq 0 ]; then
      echo '======================' 
      line_eng=$(sed -n "expr $n4 - 1p" "$filename".srt)
      time_eng=($line_eng)
      t1=${time_eng[0]}
      t2=${time_eng[2]}
      #sed "expr $n4 - 1!d" "$filename".srt >> "$filename""_eng.srt"
      #if [[ $(sed "expr $n4 - 1!d" "$filename".srt) = '-->' ]]; then sed "expr $n4 - 1!d" "$filename".srt >> "$filename"_t.txt; fi 
    fi
    jump=0 
    eng=expr $eng + 1
  fi 
  #if [[ $(sed "expr $n4 - 1!d" "$filename".srt) = '-->' ]]; then sed "expr $n4 - 1!d" "$filename".srt >> "$filename"_t.txt; fi 
  if [ $jump -eq 1 ] && [ $eng -eq 0 ] ; then
    echo "write1 $t1 $t2"
    echo "$t1 $t2" >> "$filename"_t.txt
  fi
  if [ $jump -eq 0 ] && [ $n -eq $N4 ] ; then
    echo "write2 $t1 $t2"
    echo "$t1 $t2" >> "$filename"_t.txt
  fi
  n=expr $n + 1
done < "$fullfile"


sed -i 's/,/./g' "$filename"_t.txt


ts_get_msec()
{
  read -r h m s ms <<< $(echo $1 | tr '.:' ' ' )
  h=${h#0}
  m=${m#0}
  s=${s#0}
  ms=${ms#0}
  echo $(((h60601000)+(m601000)+(s1000)+ms))
}


duration=0
i=1
while read -r start_ts stop_ts <&3; do
  i_formatted=$(printf "%04d" "$i")
  if [ $re_encode = true ] ; then
    ffmpeg -i $fullfile -ss $start_ts -to $stop_ts -c copy "$filename"_list"$i_formatted"."$extension"
  else
    ffmpeg -i $fullfile -ss $start_ts -to $stop_ts -c copy "$filename"_list"$i_formatted"."$extension"
    #ffmpeg -ss $start_ts -to $stop_ts -i $fullfile -vcodec copy -acodec copy -avoid_negative_ts make_zero "$filename"_list"$i_formatted"."$extension"
  fi


i=expr $i + 1
  START=$(ts_get_msec $start_ts)
  STOP=$(ts_get_msec $stop_ts)
  DIFF=$((STOP-START))
  duration=expr $duration + $DIFF
  echo "start $START"
  echo "stop $STOP"
  echo "diff $DIFF"
  echo "duration $duration"
done 3< "$filename"_t.txt
#https://stackoverflow.com/a/55682555/5845212
#ffmpeg will corrupt stanin, so use 3<


echo "f duration $duration"
for f in "$filename"_list*."$extension"; do echo "file '$f'" >> "$filename"_list.txt; done


if [ $re_encode = true ] ; then
  ffmpeg -f conat -safe 0 -i "$filename"_list.txt "$filename"_eng."$extension"
else
  ffmpeg -f concat -safe 0 -i "$filename"_list.txt -c copy "$filename"_eng."$extension"
fi


min=$(($duration /(601000)))
sec=$((($duration %(601000))/1000))
ms=$((($duration %(60*1000))%1000))
echo "video duration ${min}:${sec}.$ms"


if [ $clean_final = true ] ; then
  rm "$filename"_eng.srt 
  rm "$filename"_t.txt 
  rm "$filename"_list.*
fi
John Paul Qiang Chen
  • 153