Split a media file associated with an ELAN file according to some tier

split_media_by_eaf_tier.R

#!/usr/bin/env Rscript

# Split a media file associated with an ELAN file according to some tier
#
# Dependencies: ffmpeg, tidyverse (R Package)
#
# Usage:   RScript split_media_by_eaf_tier.R eaf_file tier_name
# Example: RScript split_media_by_eaf_tier.R example.eaf tier_A

# Load packages, install 'tidyverse' (it includes both xml2 and stringr in it)
require(tidyverse)
require(xml2)
require(stringr)

args <- commandArgs(trailingOnly = TRUE)

eaf_file  <- args[1]
tier_name <- args[2]

# For testing within R, uncomment as necessary
# setwd("/_sandboxes/BFoley/SoundFilePlusElanPairingsForAppDict.Individualwords")
# eaf_file  <- "KayIndivWords20170111a.eaf"
# tier_name <- "text@sp"

eaf_file <- paste0(getwd(), "/", eaf_file)

# ELAN data is stored as XML, so read in eaf data using read_xml
eaf_dat <- read_xml(eaf_file)

# Get where media file is relative to the .eaf file
med_file <- xml_find_first(eaf_dat, "//MEDIA_DESCRIPTOR") %>%
  xml_attr("RELATIVE_MEDIA_URL") %>%
  paste0(getwd(), "/", .)

# Configure output variables
out_ext <- str_extract(med_file, ".[a-z]+$")

if(!dir.exists("split")) { dir.create("split") }
out_path <- paste0(getwd(), "/split/")

# Time codes are stored separated to annotations. Need to
# grab these first, then join them to annotation data later
eaf_timecodes <- xml_find_all(eaf_dat, "//TIME_SLOT") %>%
  map_df(
    ~ xml_attrs(.) %>%
      unlist() %>%
      as.list() %>%
      data.frame(stringsAsFactors = FALSE)
  )

# Get annotations from 'tier_name' defined above
annotations_df <- xml_find_all(eaf_dat,
             paste0("TIER[@TIER_ID='",
                    tier_name,
                    "']/ANNOTATION/ALIGNABLE_ANNOTATION")
             ) %>%
  map_df(
    ~ data.frame(
      stringsAsFactors = FALSE,
      start_tid  = xml_attr(., "TIME_SLOT_REF1"),
      end_tid    = xml_attr(., "TIME_SLOT_REF2"),
      annotation = xml_child(.) %>% xml_text()
    )
  ) %>%
  left_join(rename(eaf_timecodes, start_tid = TIME_SLOT_ID, start_time = TIME_VALUE)) %>%
  left_join(rename(eaf_timecodes, end_tid = TIME_SLOT_ID, end_time = TIME_VALUE))

# Group by annotations and assign -1, -2, ... disambiguators
# just in case the same utterance occurs multiple times
annotations_df %>%
  group_by(annotation) %>%
  mutate(dis_n = 1:n(),
         max_n = max(dis_n)) %>%
  ungroup %>%
  mutate(annotation = ifelse(max_n > 1, paste0(annotation, "-", dis_n), annotation),
         start_time = as.integer(start_time) / 1000,
         end_time   = as.integer(end_time) / 1000,
         dur_secs   = end_time - start_time,
         out_file   = paste0("'", out_path, annotation, out_ext, "'")) %>%
  select(start_time, dur_secs, annotation, out_file) -> annotations_df

# Prepare for command-line execution
cmds_list <- paste0("ffmpeg -y",
                    " -i ", med_file,
                    " -ss ", annotations_df$start_time,
                    " -t ", annotations_df$dur_secs,
                    " -acodec copy ",
                    annotations_df$out_file)

# List apply commands to system()
exec_results <- lapply(cmds_list, system)

if(!all(exec_results == 0)) {
  warning("The following commands returned non-zero exits: ")
  cmds_list[unlist(exec_results)]
  quit(status = 1)
}