speech-dataset-parser 0.0.1

Library to parse speech datasets stored in a generic format based on TextGrids. A tool (CLI) for converting common datasets like LJ Speech into a generic format is included.

speech-dataset-parser

Library to parse speech datasets stored in a generic format based on TextGrids. A tool (CLI) for converting common datasets like LJ Speech into a generic format is included. Speech datasets consists of pairs of .TextGrid and .wav files. The TextGrids need to contain a tier which has each symbol separated in an interval, e.g., T|h|i|s| |i|s| |a| |t|e|x|t|.

Generic Format

The format is as follows: {Dataset name}/{Speaker name};{Speaker gender};{Speaker language}[;{Speaker accent}]/[Subfolder(s)]/{Recordings as .wav- and .TextGrid-pairs}

Example: LJ Speech/Linda Johnson;2;eng;North American/wavs/...

Speaker names can be any string (excluding ; symbols). Genders are defined via their ISO/IEC 5218 Code. Languages are defined via their ISO 639-2 Code. Accents are optional and can be any string (excluding ; symbols).

Installation

pip install speech-dataset-parser --user

Library Usage

from speech_dataset_parser import parse_dataset

entries = list(parse_dataset(folder..., grid-tier-name...))

The resulting entries list contains dataclass instances with these properties:

• symbols: Tuple[str, ...]
• intervals: Tuple[float, ...]
• symbols_language: str
• speaker_name: str
• speaker_accent: str
• speaker_gender: int
• audio_file_abs: Path
• min_time: float
• max_time: float

CLI Usage

dataset-converter-cli [-h] [-v] {convert-ljs} ...

CLI Features

• convert-ljs: convert LJ Speech dataset to a generic dataset

CLI Example

# Convert LJ Speech dataset with symbolic links to the audio files
dataset-converter-cli convert-ljs \
  "/data/datasets/LJSpeech-1.1" \
  "/tmp/ljs" \
  --tier "Symbols" \
  --symlink

Dependencies

• tqdm
• TextGrid>=1.5
• ordered_set>=4.1.0

Roadmap

• Supporting conversion of more datasets
• Adding tests

License

MIT License

Acknowledgments

Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – Project-ID 416228727 – CRC 1410

Citation

If you want to cite this repo, you can use this BibTeX-entry:

@misc{tssdp22,
  author = {Taubert, Stefan},
  title = {speech-dataset-parser},
  year = {2022},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/stefantaubert/speech-dataset-parser}}
}