waxal-wolof.py

# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Waxal Wolof Dataset."""


import csv
import os

import datasets


# TODO: Add BibTeX citation
# Find for instance the citation on arxiv or on the dataset repo/website
_CITATION = """\
@InProceedings{huggingface:dataset,
title = {A great new dataset},
author={huggingface, Inc.
},
year={2020}
}
"""

# TODO: Add a link to an official homepage for the dataset here
_HOMEPAGE = ""

_LICENSE = "Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0)"

_MODALITIES_COMBINATION = [
  ["audio", "image", "text"],
  ["audio", "text"],
  ["audio", "image"],
  ["image", "text"],
  ["audio"], 
  ["image"], 
  ["text"],
]

_URLs = {
    "train-transcriptions": "train_transcriptions.csv",
    "test-transcriptions": "test_transcriptions.csv",
    "image-files": "images.tar.gz",
    "captioned-images": "captioned_images.tar.gz",
    "audio-files": "audios.tar.gz",
    "transcribed-audio": "transcribed_audio.tar.gz"
}


class WaxalConfig(datasets.BuilderConfig):
  """BuilderConfig for Waxal dataset."""

  def __init__(self, name, version, modalities, **kwargs):
    self.modalities = modalities
    self.language = kwargs.pop("language", None)

    modalities_str = " to ".join(self.modalities)
    description = f"Waxal {modalities_str} in {self.language}"

    super(WaxalConfig, self).__init__(
        name=name,
        version=version,
        description=description,
        **kwargs,
    )


class WaxalWolof(datasets.GeneratorBasedBuilder):
  BUILDER_CONFIGS = [
      WaxalConfig(
          name="-".join(modalities),
          version=datasets.Version("1.1.0"),
          modalities=modalities,
          language="wolof",
      )
      for modalities in _MODALITIES_COMBINATION
  ]

  DEFAULT_CONFIG_NAME = "audio-text"

  def _info(self):
    features = {}

    if "audio" in self.config.modalities:
      features["audio"] = datasets.features.Audio()
      features["audio_duration"] = datasets.Value("float")
      features["participant"] = datasets.Value("int32")

    if "image" in self.config.modalities:
      features["image"] = datasets.features.Image()

    if "text" in self.config.modalities:
      features["text_annotation"] = datasets.Value("string")

    return datasets.DatasetInfo(
        description=self.config.description,
        features=datasets.Features(features),
        homepage=_HOMEPAGE,
        license=_LICENSE,
        citation=_CITATION,
    )

  @property
  def with_audio(self):
    return "audio" in self.config.modalities

  @property
  def with_image(self):
    return "image" in self.config.modalities

  @property
  def with_text(self):
    return "text" in self.config.modalities

  def _split_generators(self, dl_manager):
    audio_url_key = "transcribed-audio" if self.with_text else "audio-files"
    image_url_key = "captioned-images" if self.with_text else "image-files"

    audio_files = (
        dl_manager.download_and_extract(_URLs[audio_url_key])
        if self.with_audio
        else None
    )

    image_files = (
        dl_manager.download_and_extract(_URLs[image_url_key])
        if self.with_image
        else None
    )

    return [
        datasets.SplitGenerator(
            name=datasets.Split.TRAIN,
            gen_kwargs={
                "metadata_path": dl_manager.download(
                    _URLs["train-transcriptions"]
                ),
                "audio_files": audio_files,
                "image_files": image_files,
            },
        ),
        datasets.SplitGenerator(
            name=datasets.Split.TEST,
            gen_kwargs={
                "metadata_path": dl_manager.download(
                    _URLs["test-transcriptions"]
                ),
                "audio_files": audio_files,
                "image_files": image_files,
            },
        ),
    ]

  def _generate_examples(
      self,
      metadata_path,
      audio_files=None,
      path_to_audio="transcribed_audio",
      image_files=None,
      path_to_images="captioned_images",
  ):
    metadata = {}

    with open(metadata_path) as buf:
      reader = csv.DictReader(buf)
      for row in reader:
        del row["prompt"]  # TODO(shpotes): remove it in future versions!

        if self.with_text:
          if not row["transcription"]:
            continue

        if self.with_image:
          row["image_file_path"] = os.path.join(
              path_to_images, self.config.language, row["image_file_name"]
          )
        if self.with_audio:
          row["audio_file_path"] = os.path.join(
              path_to_audio, row["audio_file_name"]
          )  # TODO(shpotes): add lang name to the csv path.

        metadata[row["idx"]] = row

    for idx, sample in metadata.items():
      result = {}

      if self.with_audio:
        result["participant"] = sample["participant"]
        result["audio_duration"] = sample["duration"]
        audio_path = os.path.join(audio_files, sample["audio_file_path"])
        with open(audio_path, "rb") as f:
          result["audio"] = {"path": audio_path, "bytes": f.read()}

      if self.with_image:
        image_path = os.path.join(image_files, sample["image_file_path"])
        with open(image_path, "rb") as f:
          result["image"] = {"path": image_path, "bytes": f.read()}

      if self.with_text:
        result["text_annotation"] = sample["transcription"]

      yield idx, result