How to Split a YouTube Playlist Into MP3 Tracks

Today, I learned how to split a downloaded YouTube playlist into individual tracks so I could play them in VLC more easily.

I wanted to write a tutorial about it, both for my future reference and in case anyone else wants to do the same thing.

Table of contents

Requirements

The following tools are needed to follow this tutorial. I’ve given the names of packages I had to install on Arch Linux, so your package names may vary.

  • ffmpeg
  • perl (version 5)
  • yt-dlp
  • cuetools
  • python-mutagen
  • lame

Downloading the Playlist Audio

First, download the playlist with yt-dlp. The details here correspond to the playlist I was working with when first attempting this project.

yt-dlp --extract-audio --audio-format wav --audio-quality 0 \
    "https://www.youtube.com/watch?v=Hti1jmkMrnU" \
    --output "azure - nujabes playlist for the groovy yayayayy"

Flag breakdown:

  • --extract-audio saves an audio file, rather than a video one.
  • --audio-format wav specifies the audio format that should be used. I chose wav because it’s accepted by a script we use down the line.
  • --audio-quality 0 maximizes the audio quality we download.
  • --output sets the name of the saved file.

The Cuesheet Format

At this point, we have one big audio file, but we the individual tracks.

There’s only one issue: every tutorial I’ve been able to find assumes you already have a cuesheet file (with the .cue extension), which tells audio software where each track starts within the larger file.

We don’t have, one of those, though. Luckily, cuesheets are a plaintext format, and we can roll our own! There is documentation for the format on the libodraw GitHub repo, which I found very helpful. I also referenced the Wikipedia page about cue sheets, specifically the examples section.

There are three really important things I gleaned from reading these sources:

  1. The cue file is laid out with a FILE at the top, that tells it what file the tracks come from

  2. For each track in the playlist, there is a TRACK block.

    It contains info about the track, but the most important thing is the start time within the file. This is written as an INDEX 01 line.

  3. The format for INDEX is mm:ss:ff, where that third item is frames. I just left frames set to 00, since the timestamps on the YouTube video only go down to the second.

    Notice that the ‘minutes’ spot only gets two characters. That means we can’t have a track that starts later than 1 hour 40 minutes. However, you should be able to manually split your audio file into two or more 99-minute chunks, in which case your cuesheet will have multiple FILE entries. I’ve never tried this before, however.

Creating Our Own Cuesheet

With this knowledge in hand, I was able to write the following playlist.cue based on the one with the video, but tweaked to be a tiny bit more accurate for some songs.

You can write a cuesheet by hand, but as I tried to convert more playlists the novelty wore off and I got really bored.

To make it easier, I wrote a script that converts from a YouTube comment containing timestamps (the kind you often find below one of these playlists) into a cuesheet TRACK list. You can find it on my GitHub.

REM GENRE Lo-Fi
REM DATE 2024
PERFORMER Nujabes
TITLE "nujabes playlist for the groovy yayayayy"
FILE "azure - nujabes playlist for the groovy yayayayy.wav" WAVE
  TRACK 01 AUDIO
    TITLE "F.I.L.O. (Instrumental)"
    PERFORMER "Nujabes, Shing02"
    INDEX 01 00:00:00
  TRACK 02 AUDIO
    TITLE "Luv (sic)"
    PERFORMER "Nujabes, Shing02"
    INDEX 01 03:31:00
  TRACK 03 AUDIO
    TITLE "Lady Brown (instrumental)"
    PERFORMER "Nujabes, Cise Starr"
    INDEX 01 08:18:00
  TRACK 04 AUDIO
    TITLE "The Final View"
    PERFORMER "Nujabes"
    INDEX 01 11:32:00
  TRACK 05 AUDIO
    TITLE "After Hanabi ~~Listen to My Beats~~"
    PERFORMER "Nujabes"
    INDEX 01 15:14:00
  TRACK 06 AUDIO
    TITLE "battlecry"
    PERFORMER "Nujabes, Shing02"
    INDEX 01 20:31:00
  TRACK 07 AUDIO
    TITLE "Think Different"
    PERFORMER "Nujabes, Substantial"
    INDEX 01 23:52:00
  TRACK 08 AUDIO
    TITLE "Gone Are The Days"
    PERFORMER "Nujabes, Uyama Hiroto"
    INDEX 01 27:09:00
  TRACK 09 AUDIO
    TITLE "Sky Is Falling"
    PERFORMER "Nujabes, C.L. Smooth"
    INDEX 01 33:11:00
  TRACK 10 AUDIO
    TITLE "Kiss OF life"
    PERFORMER "Nujabes, Giovanca, Benny Sings"
    INDEX 01 37:53:00
  TRACK 11 AUDIO
    TITLE "Horn in the middle"
    PERFORMER "Nujabes"
    INDEX 01 42:01:00
  TRACK 12 AUDIO
    TITLE "Far Fowls"
    PERFORMER "Nujabes"
    INDEX 01 46:26:00
  TRACK 13 AUDIO
    TITLE "Summer Gypsy"
    PERFORMER "Nujabes"
    INDEX 01 50:45:00
  TRACK 14 AUDIO
    TITLE "Perfect Circle"
    PERFORMER "Nujabes"
    INDEX 01 54:46:00
  TRACK 15 AUDIO
    TITLE "Sky is Tumbling"
    PERFORMER "Nujabes, Cise Starr"
    INDEX 01 59:15:00
  TRACK 16 AUDIO
    TITLE "Kumomi"
    PERFORMER "Nujabes"
    INDEX 01 63:09:00
  TRACK 17 AUDIO
    TITLE "Luv (sic) pt2"
    PERFORMER "Nujabes, Shing02"
    INDEX 01 67:43:00
  TRACK 18 AUDIO
    TITLE "Feather"
    PERFORMER "Nujabes, Cise Starr, Akin"
    INDEX 01 70:38:00

NOTE: I chose a bad first playlist, because the timestamps that I copied from the comments don’t actually line up with the audio tracks. The above cuesheet attempts to fix those errors.

Splitting the Downloaded Audio

Now that we have a finished cuesheet, this tutorial starts looking more like others that can be found online. I’m going to see this to the end, just so I can document all the steps I ended up taking.

I used a perl script by xrgtn on GitHub to split the audio file. I initially attempted to use the shnsplit program as the Arch Wiki recommends, but I had issues where it cut tracks in the wrong places. I still don’t know what went wrong there.

./split-cue -d playlist.cue

Here, the -d flag disables the CDDB query functionality of the script. This is a feature that I didn’t require, and I didn’t want to spend the time of installing the perl module correctly.

Optional: Adding Metadata and Album Art

After the script runs, you will have a directory with an mp3 file for each track. Next, use the metadata contained in the cuesheet to flesh out each file better.

To get this script, I had to install the cuetools package.

cd '2024 - Nujabes - nujabes playlist for the groovy yayayayy'
cuetag.sh ../playlist.cue *.mp3

To add album art, you first need to find a suitable image. I went with the art linked on the YouTube playlist, https://yukoart.com/work/inq-mobile/. Save your image as image.jpg in the same directory as the individual songs.

NOTE: Try to keep your file pretty small, since its size will be added on to each track. 300x300 pixels is good.

To add the album art to a file, I found a useful command from Baeldung (https://www.baeldung.com/linux/terminal-music-add-album-art). I wrote a really simple bash loop that will add art to all the mp3 files in the current directory.

NOTE: The funkiness with the && and || lets us run a different command if ffmpeg returned a success or returned a failure.

If it succeeded (&&), we remove the temporary file. If it failed (||), we move the temporary file back to its original name to avoid messing up the directory state.

for file in *.mp3; do
  tempfile="${file}.temp"
  mv "$file" "$tempfile"
  ffmpeg -i "$tempfile" -i image.jpg -map_metadata 0 -map 0 -map 1 -acodec copy "$file" \
    && rm "$tempfile" \
    || mv "$tempfile" "$file"
done

Then, just remove the image.jpg file--it is encoded into the metadata of the mp3s, and no longer needs to exist on disk. You should be ready to listen to your music, and the metadata and album art will be available to your music program!

Pixel art of a cat with a signature next to it, reading 'almond'