tests/testthat/test-dataset-speechcommands.R

In torchaudio: R Interface to 'pytorch''s 'torchaudio'

speechcommand_dir <- system.file("SpeechCommands", package = "torchaudio")

test_that("load_speechcommands_item", {
  # load
  speechcommand_ds <- speechcommand_dataset(speechcommand_dir)
  expect_length(speechcommand_ds, 0)
  speechcommand_ds <- speechcommand_dataset(speechcommand_dir, folder_in_archive = "")
  expect_length(speechcommand_ds, 7)

  one_item <- speechcommand_ds[1]
  expect_length(one_item, 6)
  expect_tensor(one_item[[1]])

  set_audio_backend("av")
  one_item <- speechcommand_ds[1]
  expect_length(one_item, 6)
  expect_tensor(one_item[[1]])
})

test_that("normalization param works", {
  speechcommand_ds <- speechcommand_dataset(
    speechcommand_dir,
    folder_in_archive = "",
    normalization = function(x) torch::torch_max(x)
  )

  one_item <- speechcommand_ds[1]
  expect_length(one_item, 6)
  expect_tensor(one_item[[1]])
  expect_equal(as.numeric(max(one_item[[1]])), 1)
})