Google Cloud Speech-to-Text API

Mark Edmondson

2020-04-19

The Google Cloud Speech-to-Text API enables you to convert audio to text by applying neural network models in an easy to use API. The API recognizes over 80 languages and variants, to support your global user base. You can transcribe the text of users dictating to an application’s microphone or enable command-and-control through voice among many other use cases.

Read more on the Google Cloud Speech-to-Text Website

The Cloud Speech API provides audio transcription. Its accessible via the gl_speech function.

Arguments include:

Returned structure

The API returns a list of two data.frame tibbles - transcript and timings.

Access them via the returned object and $transcript and $timings

return <- gl_speech(test_audio, languageCode = "en-GB")

return$transcript
# A tibble: 1 x 2
#                                                                                                         transcript confidence
#                                                                                                              <chr>      <chr>
#1 to administer medicine to animals is frequently a very difficult matter and yet sometimes it's necessary to do so  0.9711006

return$timings
#   startTime endTime       word
#1         0s  0.100s         to
#2     0.100s  0.700s administer
#3     0.700s  0.700s   medicine
#4     0.700s  1.200s         to
# etc...

Demo for Google Cloud Speech-to-Text API

A test audio file is installed with the package which reads:

“To administer medicine to animals is frequently a very difficult matter, and yet sometimes it’s necessary to do so”

The file is sourced from the University of Southampton’s speech detection (http://www-mobile.ecs.soton.ac.uk/) group and is fairly difficult for computers to parse, as we see below:

library(googleLanguageR)
## get the sample source file
test_audio <- system.file("woman1_wb.wav", package = "googleLanguageR")

## its not perfect but...:)
gl_speech(test_audio)$transcript

## get alternative transcriptions
gl_speech(test_audio, maxAlternatives = 2L)$transcript

gl_speech(test_audio, languageCode = "en-GB")$transcript

## help it out with context for "frequently"
gl_speech(test_audio, 
            languageCode = "en-GB", 
            speechContexts = list(phrases = list("is frequently a very difficult")))$transcript

Word transcripts

The API supports timestamps on when words are recognised. These are outputted into a second data.frame that holds three entries: startTime, endTime and the word.

str(result$timings)
#'data.frame':  152 obs. of  3 variables:
# $ startTime: chr  "0s" "0.100s" "0.500s" "0.700s" ...
# $ endTime  : chr  "0.100s" "0.500s" "0.700s" "0.900s" ...
# $ word     : chr  "a" "Dream" "Within" "A" ...

result$timings
#     startTime endTime       word
#1          0s  0.100s          a
#2      0.100s  0.500s      Dream
#3      0.500s  0.700s     Within
#4      0.700s  0.900s          A
#5      0.900s      1s      Dream

Custom configurations

You can also send in other arguments which can help shape the output, such as speaker diagrization (labelling different speakers) - to use such custom configurations create a RecognitionConfig object. This can be done via R lists which are converted to JSON via library(jsonlite) and an example is shown below:

## Use a custom configuration
my_config <- list(encoding = "LINEAR16",
                  diarizationConfig = list(
                    enableSpeakerDiarization = TRUE,
                    minSpeakerCount = 2,
                    maxSpeakCount = 3
                  ))

# languageCode is required, so will be added if not in your custom config
gl_speech(my_audio, languageCode = "en-US", customConfig = my_config)

Asynchronous calls

For speech files greater than 60 seconds of if you don’t want your results straight away, set asynch = TRUE in the call to the API.

This will return an object of class "gl_speech_op" which should be used within the gl_speech_op() function to check the status of the task. If the task is finished, then it will return an object the same form as the non-asynchronous case.

async <- gl_speech(test_audio, asynch = TRUE)
async
## Send to gl_speech_op() for status
## 4625920921526393240

result <- gl_speech_op(async)