multilingual_tts_ljspeech.py

"""Multilingual TTS Dataset in LJSpeech Format"""

import csv
import json
import os
import zipfile
from pathlib import Path

import datasets
from datasets import Features, Value, Audio


_CITATION = """@dataset{multilingual_tts_ljspeech,
  title={Multilingual TTS Dataset in LJSpeech Format},
  year={2024},
  note={English: LibriTTS-R (CC BY 4.0), Chinese: AISHELL-3 (Apache 2.0)}
}
"""

_DESCRIPTION = """A high-quality multilingual Text-to-Speech dataset in LJSpeech format, 
containing English and Chinese speech data suitable for commercial use.

This dataset combines:
- English: LibriTTS-R (~49 hours, 247 speakers, 32K utterances)
- Chinese: AISHELL-3 (~49 hours, 174 speakers, 63K utterances)

Total: ~97 hours, 421 speakers, 95K utterances
All audio normalized to 22050Hz, 16-bit, mono WAV format.
"""

_HOMEPAGE = "https://huggingface.co/datasets/ayousanz/multi-dataset-v2"
_LICENSE = "Mixed: CC BY 4.0 (English), Apache 2.0 (Chinese)"


class MultilingualTTSLJSpeech(datasets.GeneratorBasedBuilder):
    """Multilingual TTS Dataset in LJSpeech Format"""

    VERSION = datasets.Version("1.0.0")

    def _info(self):
        features = Features({
            "audio_id": Value("string"),
            "audio": Audio(sampling_rate=22050),
            "transcription": Value("string"),
            "normalized_text": Value("string"),
            "speaker_id": Value("string"),
            "language": Value("string"),
        })

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

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        # Download files with clean structure
        files_to_download = {
            # Metadata files
            "train_csv": "metadata/train.csv",
            "validation_csv": "metadata/validation.csv", 
            "test_csv": "metadata/test.csv",
            
            # Audio ZIP files
            "train_en_zip": "audio/train_english.zip",
            "train_zh_zip": "audio/train_chinese.zip",
            "validation_en_zip": "audio/validation_english.zip",
            "validation_zh_zip": "audio/validation_chinese.zip",
            "test_en_zip": "audio/test_english.zip",
            "test_zh_zip": "audio/test_chinese.zip",
        }
        
        downloaded_files = dl_manager.download(files_to_download)
        
        # Extract audio ZIP files
        extracted_dirs = {}
        for split in ["train", "validation", "test"]:
            for lang in ["en", "zh"]:
                lang_name = "english" if lang == "en" else "chinese"
                key = f"{split}_{lang}_zip"
                if key in downloaded_files:
                    extracted_dirs[f"{split}_{lang}"] = dl_manager.extract(downloaded_files[key])
        
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "metadata_path": downloaded_files["train_csv"],
                    "extracted_dirs": extracted_dirs,
                    "split": "train",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "metadata_path": downloaded_files["validation_csv"],
                    "extracted_dirs": extracted_dirs,
                    "split": "validation",
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "metadata_path": downloaded_files["test_csv"],
                    "extracted_dirs": extracted_dirs,
                    "split": "test",
                },
            ),
        ]

    def _generate_examples(self, metadata_path, extracted_dirs, split):
        """Yields examples."""
        with open(metadata_path, encoding="utf-8") as f:
            reader = csv.DictReader(f, delimiter="|")
            
            for idx, row in enumerate(reader):
                audio_id = row["audio_id"]
                language = row["language"]
                
                # Find the audio file in extracted directories
                audio_dir_key = f"{split}_{language}"
                if audio_dir_key in extracted_dirs:
                    audio_path = Path(extracted_dirs[audio_dir_key]) / split / language / "wavs" / f"{audio_id}.wav"
                    
                    if audio_path.exists():
                        yield idx, {
                            "audio_id": audio_id,
                            "audio": str(audio_path),
                            "transcription": row["transcription"],
                            "normalized_text": row["normalized_text"],
                            "speaker_id": row["speaker_id"],
                            "language": language,
                        }