vakya

CLI tool to extract audio files and metadata from Parquet datasets into a structured, portable format.

Installation

pip install vakya

Quick Start

1. Inspect a Parquet file to see its columns:

   vakya schema data/train-00000.parquet

2. Validate your config (or use defaults):

   vakya validate-config my_config.yaml

3. Extract everything:

   vakya extract ./my_dataset --config my_config.yaml

Configuration (YAML)

All behavior is controlled via a YAML config file.

# vakya_config.yaml

# Column in the Parquet file that contains audio data (dict with "bytes").
audio_column: audio

# Output audio file extension.
audio_format: flac

# Column used for audio filename.
filename_column: fname

# List of columns to include in the metadata object.
metadata_columns:
  - fname
  - text
  - duration
  - gender
  - speaker_id
  - lang

# Name of the master JSONL file. Set to null to disable.
master_jsonl: master.jsonl

# Number of parallel workers (1 to CPU count).
workers: 1

# Whether to overwrite existing audio files.
overwrite: false

# Log level: DEBUG | INFO | WARNING | ERROR
log_level: INFO

CLI Commands

• vakya extract ROOT_DIR: Recursively finds Parquet files and extracts them.
• vakya validate-config CONFIG_PATH: Checks if your YAML config is valid.
• vakya schema PARQUET_FILE: Prints the Arrow schema and row count of a Parquet file.

Output Format

root_dataset/
├── hindi/
│   ├── train-00000.parquet
│   ├── hindi_audio/              ← Audio files here
│   │   ├── sample_001.flac
│   │   └── sample_002.flac
│   └── hindi.jsonl               ← Metadata for this folder
├── master.jsonl                  ← Aggregated metadata
└── config.yaml

Each line in the .jsonl files:

{
  "audio_path": "hindi/hindi_audio/sample_001.flac",
  "metadata": {
    "fname": "sample_001",
    "text": "Hello world",
    "duration": 3.45,
    "gender": "female"
  }
}

Development

Install development dependencies:

pip install -e ".[dev]"

Run tests:

pytest

Run linting (Ruff):

ruff check src/ tests/

Run type checking (Mypy):

mypy src/ tests/

CI/CD and Branching

• Main Branch: master
• CI: Every push/PR to master runs linting, type checking, and tests.
• Publishing: Automatic publishing to PyPI on version tags (v*.*.*) after tests pass. Requires GitHub environment approval.

License

MIT