vps_clustering_benchmark.py

import os
import json
import datasets
import pandas as pd
import numpy as np
from typing import List
from tqdm import tqdm

logger = datasets.logging.get_logger(__name__)

_DESCRIPTION = """\
This dataset consists of a small sample of audio clips with annotated
speaker identities, their age and gender, diarization-based speech segments,
and transcription. The dataset is intended for the benchmarking of VBI-core.
"""

_HF_REPO_PATH = "https://huggingface.co/datasets/conversy/vps_clustering_benchmark/resolve/main/voice_prints.pkl"


class VPClusteringBenchmarkConfig(datasets.BuilderConfig):
    """BuilderConfig for Conversy Benchmark."""

    def __init__(self, name, version, **kwargs):
        """BuilderConfig for Conversy Benchmark.
        Args:
          **kwargs: keyword arguments forwarded to super.
        """
        self.name = name
        self.version = version
        self.features = kwargs.pop("features", None)
        self.description = kwargs.pop("description", None)
        self.data_url = kwargs.pop("data_url", None)
        self.nb_data_shards = kwargs.pop("nb_data_shards", None)

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


class VPClusteringBenchmark(datasets.GeneratorBasedBuilder):
    """Conversy benchmark"""
    VERSION = datasets.Version("1.0.0")
    BUILDER_CONFIGS = [
        VPClusteringBenchmarkConfig(
                name="VPClusteringBenchmark",
                version=VERSION,
                description="Conversy Benchmark for ML models evaluation",
                features=["segment_id", "filename", "speaker", "duration", "vp",
                          "segment_clean"],
                data_url=_HF_REPO_PATH,
                nb_data_shards=1)
    ]

    def _info(self):
        description = (
            "Voice Print Clustering Benchmark"
        )
        features = datasets.Features(
            {
                "segment_id": datasets.Value("string"),
                "filename": datasets.Value("string"),
                "speaker": datasets.Value("string"),
                "duration": datasets.Value("float32"),
                "segment_clean": datasets.Value("bool"),
                "vp": datasets.Sequence(datasets.Value("float32"))
            })
        return datasets.DatasetInfo(
            description=description,
            features=features,
            supervised_keys=None,
            version=self.config.version
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        data_url = self.config.data_url
        downloaded_file = dl_manager.download_and_extract(data_url)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"file_path": downloaded_file},
            ),
        ]

    def _generate_examples(self, file_path):
        """Yields examples."""
        df = pd.read_pickle(file_path)

        for idx, row in df.iterrows():
            yield idx, {
                "segment_id": row["segment_id"],
                "filename": row["filename"],
                "speaker": row["speaker"],
                "duration": row["duration"],
                "segment_clean": row["segment_clean"],
                "vp": row["vp"].tolist() if isinstance(row["vp"], np.ndarray) else row["vp"]
            }