I wanted to download videos (or more precisely I was only interested in the audio) from youtube. Unfortunately, some videos had more intro and outro than just the music I wanted. Fortunately ffmpeg in combination with youtube-dl and some shell scripting is the key to solve the problem:

#!/bin/bash#!/bin/bash
# massively inspired from https://github.com/rg3/youtube-dl/issues/622 & man youtube-dl & man ffmpeg & google 
# loop over download-links.csv, with a semicolon as delimiter. 
# Columns with CAPITALS are actually needed, others only to calculate start (SS) and duration (T)

IFS=";"
while read TITLE URL start end START_TIME END_TIME
do 
  if [[ "$URL" == "http"* ]]; then 
    echo "### Downloading $TITLE:" 
    FILENAME="$TITLE" 
    EXT="mp3" 
    AUTHOR="artist" 
    ALBUM="album" 
    # download audio only from START_TIME for duration END_TIME 
    # also write id3 tags. Currently aritst and album is hardcoded 
    ffmpeg \ 
      -n \ 
      -nostdin \ 
      -i $(youtube-dl -f 22 --get-url $URL) \ 
      -ss $START_TIME \ 
      -t $END_TIME \ 
      -c:a libmp3lame -qscale:a 2 \ 
      -metadata title="$TITLE" \ 
      -metadata author="$AUTHOR" \ 
      -metadata album="$ALBUM" \ 
      "$FILENAME.$EXT" 
      # finally check the file and id3 tag 
      ffprobe "$FILENAME.$EXT" 2>&1 | grep -A90 'Metadata:' 
  fi
done < download-links.csv

# alternatively download video
#ffmpeg -ss $START_TIME -i $(youtube-dl -f 22 --get-url $URL) -t $END_TIME -c:v copy -c:a copy "$FILENAME.$EXT"

All you need is to create a csv-file following the following structure: