bam-asr-all.py

"""
Copyright 2025 RobotsMali AI4D Lab.

Licensed under the Creative Commons Attribution 4.0 International License (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

https://creativecommons.org/licenses/by/4.0/

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.
"""

import csv
import datasets
from datasets import Split, SplitGenerator

# -----------------------
# 1. Basic meta-infos
# -----------------------
_CITATION = """\
@inproceedings{bam_asr_all_2025,
  title={Bam-ASR-All Audio Dataset},
  author={RobotsMali AI4D Lab},
  year={2025},
  publisher={Hugging Face}
}
"""

_DESCRIPTION = """
The **Bam-ASR-All** dataset is a combined Bambara speech dataset featuring multiple subsets:
- Oza-Mali-Pense
- Jeli-ASR
- RT-Data-Collection
All subsets contain audio samples in Bambara along with transcriptions and (potentially) 
French translations.
"""

_HOMEPAGE = "https://huggingface.co/datasets/RobotsMali/bam-asr-all"
_LICENSE = "CC-BY-4.0"
_VERSION = datasets.Version("1.0.0")

# NOTE: No trailing slash here
_BASE_URL = "https://huggingface.co/datasets/RobotsMali/bam-asr-all/resolve/main"

# -----------------------
# 2. Config + Builder
# -----------------------
class BamASRAllConfig(datasets.BuilderConfig):
    """BuilderConfig for different subsets of Bam-ASR-All dataset."""

class BamASRAll(datasets.GeneratorBasedBuilder):
    """
    This class defines how to load and parse the Bam-ASR-All dataset 
    from metadata.csv + audio files on the Hub.
    """

    # 2a. Define your subsets (configs)
    BUILDER_CONFIGS = [
        BamASRAllConfig(
            name="oza-mali-pense",
            version=_VERSION,
            description="Load only the Oza-Mali-Pense subset (files under oza-mali-pense/).",
        ),
        BamASRAllConfig(
            name="jeli-asr",
            version=_VERSION,
            description="Load only the Jeli-ASR subset (files under jeli-asr/).",
        ),
        BamASRAllConfig(
            name="rt-data-collection",
            version=_VERSION,
            description="Load only the RT-Data-Collection subset (files under rt-data-collection/).",
        ),
        # The "combined" option for everything can also be done
        BamASRAllConfig(
            name="bam-asr-all",  # The dataset's default name
            version=_VERSION,
            description="Combine oza-mali-pense, jeli-asr, and rt-data-collection (all rows).",
        ),
    ]

    # 2b. Default subset name if none specified
    DEFAULT_CONFIG_NAME = "bam-asr-all"

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "audio": datasets.Audio(sampling_rate=16_000),
                    "duration": datasets.Value("float32"),
                    "bam": datasets.Value("string"),
                    "french": datasets.Value("string"),
                }
            ),
            supervised_keys=None,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    # -----------------------
    # 3. Splits
    # -----------------------
    def _split_generators(self, dl_manager):
        """
        1) Download 'metadata.csv' from the Hub by specifying its raw URL.
        2) We'll then yield two splits (TRAIN, TEST) by reading that CSV 
           and filtering rows by '/train/' or '/test/' in file paths.
        """
        metadata_url = f"{_BASE_URL}/metadata.csv"
        local_metadata_path = dl_manager.download(metadata_url)

        return [
            SplitGenerator(
                name=Split.TRAIN,
                gen_kwargs={
                    "metadata_path": local_metadata_path,
                    "split": "train",
                    "dl_manager": dl_manager,
                },
            ),
            SplitGenerator(
                name=Split.TEST,
                gen_kwargs={
                    "metadata_path": local_metadata_path,
                    "split": "test",
                    "dl_manager": dl_manager,
                },
            ),
        ]

    # -----------------------
    # 4. Generate examples
    # -----------------------
    def _generate_examples(self, metadata_path, split, dl_manager):
        """
        Read metadata.csv row-by-row, filter by:
          - the config name (oza-mali-pense, jeli-asr, rt-data-collection, or all)
          - 'train' vs 'test' in file path
        Then download each audio file from the Hub, yield local path + metadata.
        """
        audios_to_download = []
        metadata_dict = {}

        with open(metadata_path, "r", encoding="utf-8") as f:
            reader = csv.DictReader(f)
            for idx, row in enumerate(reader):
                file_path = row["file_name"]  # e.g. "jeli-asr/train/.../some.wav"

                # Filter by config name
                if self.config.name == "oza-mali-pense":
                    if "oza-mali-pense/" not in file_path:
                        continue
                elif self.config.name == "jeli-asr":
                    if "jeli-asr/" not in file_path:
                        continue
                elif self.config.name == "rt-data-collection":
                    if "rt-data-collection/" not in file_path:
                        continue
                elif self.config.name == "bam-asr-all":
                    # Keep all rows
                    pass

                # Filter by split (train/test)
                if split == "train" and "/train/" not in file_path:
                    continue
                if split == "test" and "/test/" not in file_path:
                    continue

                # Build the raw URL for this audio file
                audio_url = f"{_BASE_URL}/{file_path}"
                audios_to_download.append(audio_url)

                # Store minimal metadata in a dictionary
                metadata_dict[audio_url] = {
                    "duration": float(row["duration"]),
                    "bam": row["bam"],
                    "french": row["french"],
                }

        # Download the audios. dl_manager returns the local paths in the cache.
        local_audio_paths = dl_manager.download(audios_to_download)
        for idx, audio_url in enumerate(audios_to_download):
            local_audio_path = local_audio_paths[idx]
            yield idx, {
                "audio": local_audio_path,  # local path for datasets.Audio
                "duration": metadata_dict[audio_url]["duration"],
                "bam": metadata_dict[audio_url]["bam"],
                "french": metadata_dict[audio_url]["french"],
            }