speech_to_text_with_audio.whisper.R

speech_to_text <- function(audio, 
                           output_name = "output",
                           model_path = "ggml-large-v3.bin"){
  library(tidyverse)
  library(audio.whisper)
  
# convert to the format specs ---------------------------------------------
  tmp <- tempdir()
  str_glue("ffmpeg -i {audio} -ar 16000 -ac 1 -c:a pcm_s16le {tmp}/{output_name}.wav") |> 
    system()
  
# load and run model ------------------------------------------------------
  model <- whisper(model_path)  
  result <- predict(model,
                    str_glue("{tmp}/{output_name}.wav"),
                    language = "ru")

# write down results ------------------------------------------------------
  result$data |> 
    as.data.frame() |> 
    write_csv(str_glue("{output_name}.csv"))
  result$params
  print(result$timing)
}

speech_to_text(audio = "test.wav", output_name = "test")