MioTTS/tools/convert_to_ljspeech_csv.py

import argparse
import csv
import os
from typing import Iterable, TextIO


def _iter_rows_with_header(
    fin: TextIO,
    file_col: str,
    text_col: str,
) -> Iterable[tuple[str, str]]:
    reader = csv.DictReader(fin)
    if reader.fieldnames is None or file_col not in reader.fieldnames or text_col not in reader.fieldnames:
        msg = (
            "Input CSV must have a header with at least the columns "
            f"{file_col!r} and {text_col!r}. Found: {reader.fieldnames!r}"
        )
        raise ValueError(msg)

    for row in reader:
        file_name = (row.get(file_col) or "").strip()
        text = (row.get(text_col) or "").strip()
        if not file_name or not text:
            continue
        yield file_name, text


def _convert(
    input_path: str,
    output_path: str,
    file_col: str,
    text_col: str,
) -> None:
    with open(input_path, encoding="utf-8-sig", newline="") as fin, open(
        output_path,
        "w",
        encoding="utf-8",
        newline="\n",
    ) as fout:
        for file_name, text in _iter_rows_with_header(fin, file_col=file_col, text_col=text_col):
            base = os.path.basename(file_name)
            file_id, _ = os.path.splitext(base)
            file_id = file_id.strip()
            if not file_id:
                continue

            normalized_text = text
            original_text = text
            fout.write(f"{file_id}|{normalized_text}|{original_text}\n")


def main() -> None:
    parser = argparse.ArgumentParser(
        description=(
            "Convert a simple CSV (e.g. file_name,text) into an LJSpeech-style metadata.csv "
            "with lines of the form '<file_id>|<normalized_text>|<original_text>'."
        )
    )
    parser.add_argument("--input", required=True, help="Path to the input CSV file.")
    parser.add_argument("--output", required=True, help="Path to write the LJSpeech-style metadata.csv.")
    parser.add_argument(
        "--file-col",
        default="file_name",
        help="Name of the column containing the audio file path (default: file_name).",
    )
    parser.add_argument(
        "--text-col",
        default="text",
        help="Name of the column containing the transcript text (default: text).",
    )
    args = parser.parse_args()

    input_path = os.path.abspath(args.input)
    output_path = os.path.abspath(args.output)

    if not os.path.isfile(input_path):
        msg = f"input CSV file not found: {input_path}"
        raise FileNotFoundError(msg)

    os.makedirs(os.path.dirname(output_path) or ".", exist_ok=True)

    print(f"Reading CSV : {input_path}")
    print(f"Writing LJS : {output_path}")

    _convert(
        input_path=input_path,
        output_path=output_path,
        file_col=args.file_col,
        text_col=args.text_col,
    )


if __name__ == "__main__":
    main()