Extract and prettify YouTube transcripts

parse_youtube_transcript.py

# -*- coding: utf-8 -*-
# @Author: Isaac Pei
# @Date:   2020-09-28 18:11:14
# @Last Modified by:   Isaac Pei
# @Last Modified time: 2021-05-29 13:49:12

## Take the vtt file as input, to generate the transcript text, 1 per line
# --
# -- There is an online tool doing similar work: https://hierogly.ph/

import sys
import re

vtt_input = sys.argv[1]
previous_line = "" # Define previous line, only when line is different then print

with open(vtt_input) as f:
    for l in f:
        if "align:start" in l:
            pass
        elif "<c>" in l:
            # for those <c> </c> marked split worded lines 
            # there are often repeated full line in the next few lines
            pass
        elif l.strip():
            # Reach below only when line is not empty
            line=re.sub(r"<.*>", " ", l.strip())
            if line != previous_line:
                print(line)
            previous_line = line

yt_dl_transcript.sh

#!/usr/bin/env bash

# DESCRIPTION
# Given the input url, download the transcipt from url ($1)

# USAGE
# yt_transcript_dl.sh $URL
# e.g. yt_transcript_dl.sh https://youtu.be/LSvJX2pJxQ

# dependencies for MacOS, comment these out if not needed
# needs youtube-dl
brew install youtube-dl
# needs jq
brew install jq
# needs coreutils
brew install coreutils

# The following command is used to download the youtube transcript
url="$1"

mkdir -p input/
mkdir -p output/

# getting title
title=$(youtube-dl -f mp4 -o '%(id)s.%(ext)s' --print-json --no-warnings --skip-download "$url" | jq -r .title)
title=$(echo "$title" | sed 's/ //g')
# downloading transcript
youtube-dl --write-auto-sub --output "input/transcript" --skip-download "$url"
# cleaning up
python ./parse_youtube_transcript.py input/transcript.en.vtt > output/"$title"-transcript.txt

rm -rf ./input

echo "The transcript is available at:" $(find . -iname "$title*" -exec realpath {} \;)