muschellij2 badge:
Ari is an R package designed to help you make videos from plain text files. Ari uses Amazon Polly to convert your text into speech. You can then provide images or a set of HTML slides which Ari will narrate based on a script you provide. Ari uses FFmpeg to stitch together audio and images.
install.packages("ari")You can also install the development version of Ari from GitHub with:
# install.packages("devtools")
devtools::install_github("seankross/ari@dev")You also need to make sure you have FFmpeg version 3.2.4 or higher installed on your system.
First create an HTML slide presentation based in R Markdown using a package
like rmarkdown or
xaringan. To see an example presentation
enter browseURL(ari_example("ari_intro.html")) into the R console. For every
slide in the package you should write some text that will be read while the
slide is being shown. You can do this in a separate Markdown file (see
file.show(ari_example("ari_intro_script.md"))) or you can use HTML comments
to put narration right in your .Rmd file (see
file.show(ari_example("ari_comments.Rmd"))). Make sure to knit your .Rmd
file into the HTML slides you want to be turned into a video.
Once you have finished your script and slides install the
aws.polly package. You can find a
guide for quickly setting up R to use Amazon Web Services
here.
Run aws.polly::list_voices() to make sure your keys are working (this
function should return a data frame). Once you've set up your access keys you
can start using Ari.
These examples make use of the ari_example() function. In order to view the
files mentioned here you should use file.show(ari_example("[file name]")).
You can watch an example of a video produced by Ari
here.
library(ari)
# First set up your AWS keys
Sys.setenv("AWS_ACCESS_KEY_ID" = "EA6TDV7ASDE9TL2WI6RJ",
"AWS_SECRET_ACCESS_KEY" = "OSnwITbMzcAwvHfYDEmk10khb3g82j04Wj8Va4AA",
"AWS_DEFAULT_REGION" = "us-east-2")
# Create a video from a Markdown file and slides
ari_narrate(
ari_example("ari_intro_script.md"),
ari_example("ari_intro.html"),
voice = "Joey")
# Create a video from an R Markdown file with comments and slides
ari_narrate(
ari_example("ari_comments.Rmd"),
ari_example("ari_intro.html"),
voice = "Kendra")
# Create a video from images and strings
ari_spin(
ari_example(c("mab1.png", "mab2.png")),
c("This is a graph.", "This is another graph"),
voice = "Joanna")
# Create a video from images and Waves
library(tuneR)
ari_stitch(
ari_example(c("mab1.png", "mab2.png")),
list(noise(), noise()))Some html slides take a bit to render on webshot, and can be dark gray instead of white. If you change the delay argument in ari_narrate, passed to webshot, this can resolve some issues, but may take a bit longer to run. Also, using capture_method = "vectorized" is faster, but may have some issues, so run with capture_method = "iterative" if this is the case as so:
ari_narrate(
ari_example("ari_comments.Rmd"),
ari_example("ari_intro.html"),
voice = "Kendra",
delay = 0.5,
capture_method = "iterative")Creating videos from plain text has some significant advantages:
- Video content can be version controlled with Git and GitHub - after all it's just plain text!
- Videos with explicit narration are more accessible to students. We don't have to rely on YouTube's often faulty captioning algorithm.
- Scripts can be automatically translated into other languages with services like the Google Translation API and Amazon Polly can speak languages other than English. This means you can write a lecture once and generate slides and videos in multiple languages.
At the Johns Hopkins Data Science Lab we rapidly develop highly technical content about the latest libraries and technologies available to data scientists. Video production requires a significant time investment and APIs are always changing. If the interface to a software library changes it's particularly arduous to re-record an entire lecture because some function arguments changed. By using Ari we hope to be able to rapidly create and update video content.