#Videogrep
Videogrep lets you make supercuts based on the spoken word content of videos. In order for it to work you need to have either a subtitle file (ending with .srt) or a transcription.
Documentation can be found at: http://antiboredom.github.io/videogrep/
Videogrep can be used as a desktop app for Mac, and a also as a command line program. The desktop app is easier to use, but less powerful than the command line program.
pip install videogrep
(you may also need to add sudo before the pip)
If this doesn't work you probably don't have pip installed. To get install it, first download get-pip.py. Then run:
python get-pip.py
For a single video:
videogrep --input my_video.mp4 --search 'some word or phrase' --output my_new_video.mp4
For multiple videos:
videogrep --input *.mp4 --search 'some word or phrase' --output my_new_video.mp4
A regular expression is a special syntax for doing advanced text search. You add special characters to your search to specify different options. Here's a good intro. And here's another one focused on Python, by Allison Parrish.
Some quick examples useful for videogrep:
Trump would match with a line that contains the word "trump" anywhere in it.
Trump|Clinton would match with a line containing either the word "Trump" or "Clinton"
^Trump would match with a line starting with the word "Trump".
Trump$ would match with a line ending with the word "Trump".
You can just put regular expressions into videogrep like a normal search:
videogrep --input myvideo.mp4 --search '^why|^because' --output questions_and_answers.mp4
You can add the --search-type flag to search for grammatical patterns or categories of words.
To search for grammatical patterns, add --search-type pos to the command, and then search for parts of speech tags instead of words.
For example, "NN" means singular noun, "VB" means basic verb, and "VBG" means gerund (a verb ending in "ing"). These are known as Penn Treeband tags, and you can see a full list of them here
So, to get all the lines of speech containing gerunds:
videogrep --input myvideo.mp4 --search-type pos --search 'VBG' --output gerunds.mp4
You could also combine this with the regular expression search. So, to find all lines starting with a gerund, just add a carat character before "VBG":
videogrep --input myvideo.mp4 --search-type pos --search '^VBG' --output gerunds.mp4
You can also search for "hypernyms" or categories of words. For example, to search for all the words that are 'liquids' you can enter:
videogrep --input myvideo.mp4 --search-type hyper --search 'liquids' --output liquids.mp4
Transcribing videos is useful when a subtitle file isn't available, or you want to make a supercut from individual words rather than phrases.
You can transcribe videos using the desktop app. If you want to do it on the command line, first install pocketsphinx:
brew tap watsonbox/cmu-sphinx
brew install --HEAD watsonbox/cmu-sphinx/cmu-sphinxbase
brew install --HEAD watsonbox/cmu-sphinx/cmu-sphinxtrain # optional
brew install --HEAD watsonbox/cmu-sphinx/cmu-pocketsphinx
To transcribe a video, just add the --transcribe flag like so:
videogrep --input myvideo.mp4 --transcribe
It'll take a while. When it's done, you can tell videogrep to use the transcription by adding the --use-transcript flag:
videogrep --input myvideo.mp4 --use-transcript --search "trump"
The --search-type flag will work with transcribed videos also.
You can also use videogrep from a Python script rather than the command line.
Here's an example of how to automatically create a supercut of the most common phrase in any video file. I used a similar process to create this twitter bot.
import sys
from pattern.en import ngrams
import videogrep
from collections import Counter
videofile = sys.argv[1]
subtitlefile = videofile.replace('.mp4', '.srt')
lines = open(subtitlefile).read()
grams = ngrams(lines, n=3)
most_common = Counter(grams).most_common(1)
phrase = most_common[0][0]
phrase = ' '.join(phrase)
print phrase
outputfile = videofile + '.most_common.mp4'
videogrep.videogrep([videofile], outputfile, phrase, 're')