Automated Translations with R and Google Translator API
Automated Translations of Videos Closed Captions with R and Google Translator API
Automated Translations of Videos Closed Captions with R and Google Translator API
Recent development of cloud services allowing to easily integrate translations ‘speach-to-text’. In particular all videos for my online courses were captioned by the learning platform. I have manually checked english auto-generated subtitles and decided to study ways how can I translate them to other languages. For example I would be able to delight my students with option to read subtitles. This may help them to better catch idea from voice on the video.
This post explains the idea to use R and Google Translate API to completely automate process and even introducing dedicated R package for that…
What are the options?
First idea is obvious! Open the file, copy and paste to Google Translator Webpage, translate and back… This is however something undersireable:
# In each course: ~ 23 videos -> 10 languages
23 * 10
## [1] 230
Second idea was to automate the process. Bad news that I am not a programmer. The only remote chance was to try using R which may allow something more realistic.
Solving for one
I started to read file with Closed Captions into R:
library(magrittr)
# read file -> it will be a dataframe
read.delim("L0.vtt", stringsAsFactors = F) %>% head()
## WEBVTT
## 1 0
## 2 00:03.330 --> 00:07.020
## 3 Hello! We all know about existence of Google translator.
## 4 1
## 5 00:07.230 --> 00:08.240
## 6 This is free.
Next step was to perform some manipulations with text strings. In particular Google charge for each translated symbol, hence sending time-stamps would be inefficient…
library(tidyverse)
# read file -> it will be a dataframe
t <- read.delim("L0.vtt", stringsAsFactors = F)
# extract logical vector indicating which rows containing timestamps
x <- t %>%
# detect rows with date time (those for not translate)
apply(MARGIN = 1, str_detect, pattern = "-->")
# extract only rows containing text (e.g. not containing timestamps)
txt <- subset.data.frame(t, !x)
# extract only time stamps
tst <- subset.data.frame(t, x)
head(txt)
## WEBVTT
## 1 0
## 3 Hello! We all know about existence of Google translator.
## 4 1
## 6 This is free.
## 7 2
## 9 Millions of people using it every day.
This dataframe txt can now be translated… good news is that someone already thought about translation with Google Cloud Platform and Translate API from R.
Automating translation has a cost. Hence anyone who want to use this service should pay. Official cost is 20$ for 1000000 symbols translated. Great news is that Google is giving $300 and 12 months free trials. API key then can be generated from the Google Cloud Platform.
I usually encrypt any of the credentials used in the R scripts so here I will demonstrate how I will translate my text extract to German (language I am currently study)
library(translateR)
library(openssl)
# get back our encrypted API key
out <- read_rds("api_key.enc.rds")
api_key <- decrypt_envelope(out$data, out$iv, out$session, "C:/Users/fxtrams/.ssh/id_api", password = "") %>%
unserialize()
# translate object txt or file in R
# Google, translate column in dataset
google.dataset.out <- translate(dataset = txt,
content.field = 'WEBVTT',
google.api.key = api_key,
source.lang = 'en',
target.lang = 'de')
# extract only new column
trsltd <- google.dataset.out %>% select(translatedContent)
# give original name
colnames(trsltd) <- "WEBVTT"
head(trsltd)
## WEBVTT
## 1 0
## 3 Hallo! Wir alle wissen über die Existenz von Google Übersetzer.
## 4 1
## 6 Das ist gratis.
## 7 2
## 9 Millionen von Menschen nutzen es jeden Tag.
Next I will rejoin timestamps with translated text and order this dataframe by rownames
# bind rows with original timestamps
abc <- rbind(tst, trsltd)
# order this file back again
bcd <- abc[ order(as.numeric(row.names(abc))), ] %>% as.character() %>% as.data.frame()
head(bcd)
## .
## 1 0
## 2 00:03.330 --> 00:07.020
## 3 Hallo! Wir alle wissen über die Existenz von Google Übersetzer.
## 4 1
## 5 00:07.230 --> 00:08.240
## 6 Das ist gratis.
I should add empty row otherwise it will not work and write to Write to the file…
# return original name
colnames(bcd) <- "WEBVTT"
bcd <- as.tibble(bcd)
# add one row
bcd2 <- add_row(bcd, WEBVTT = "", .before = 1)
# write this file back :_)
write.table(bcd2, "translated.vtt",quote = F, row.names = F, fileEncoding = "UTF-8")
Function and even R Package
All the code manipulations were nicely placed into the R function. I went a bit further by learning how to create an R package…
# installing the R package
devtools::install_github("vzhomeexperiments/translateVTT")
After that I can just give a file path to the closed caption file and it will does everything by itself:
# load R package
library(translateVTT)
# send VTT file for translation and writing it back
translateVTT(fileName = "L0.vtt",
sourceLang = "en",
destLang = "de",
apikey = api_key)
## [1] "LC_COLLATE=English_United Kingdom.1252;LC_CTYPE=English_United Kingdom.1252;LC_MONETARY=English_United Kingdom.1252;LC_NUMERIC=C;LC_TIME=English_United Kingdom.1252"
read.delim("L0.vttnl.vtt") %>% head()
## WEBVTT
## 1 0
## 2 00:03.330 --> 00:07.020
## 3 Hallo! We weten allemaal over het bestaan <U+200B><U+200B>van Google vertaler.
## 4 1
## 5 00:07.230 --> 00:08.240
## 6 Dit is gratis.
More automation?
Real usecase may entail keeping all my files to translate in one folder. I can read the names of those files and store them in a vector.
# make a list of files to translate
filesToTranslate <-list.files("C:/Users/fxtrams/Downloads/", pattern="*.vtt", full.names=TRUE)
head(filesToTranslate)
## [1] "C:/Users/fxtrams/Downloads/L0.vtt" "C:/Users/fxtrams/Downloads/L1.vtt"
## [3] "C:/Users/fxtrams/Downloads/L2.vtt" "C:/Users/fxtrams/Downloads/L3.vtt"
## [5] "C:/Users/fxtrams/Downloads/L4.vtt" "C:/Users/fxtrams/Downloads/L5.vtt"
I will also be able to setup vector with languages symbols:
# make a list of languages
languages <- c("fr", "tr", "it", "id", "pt", "es", "ms", "de")
Finally basic for loop in R will do the entire job:
# for loop
for (FILE in filesToTranslate) {
# for loop for languages
for (LANG in languages) {
# translation
translateVTT(fileName = FILE, sourceLang = "en", destLang = LANG, apikey = api_key)
}
}
Conclusion!
Entire exercise lasted 1.26 hours and consumed 18.37$. It has translated my 23 files to 8 different languages including Chinese… Things that would be otherwise impossible to accomplish without Google Translate API