append_to_dataset.py

import argparse
import os
import re
import ass
import pydub
import json
import string

def break_line(line: str):
    pattern = re.compile(r"^\[(\d+.\d+)\](.*)\[(\d+.\d+)\]$")
    match = pattern.match(line)
    start_time = match.group(1)
    text = match.group(2)
    end_time = match.group(3)
    return start_time, text.strip(), end_time, float(start_time), float(end_time)

def to_ascii_filename(text, fallback="file"):
  """Converts a string to a valid filename.

  Args:
    text: The string to convert.

  Returns:
    A lowercase filename consisting of letters, numbers, underscores, hyphens,
    and dots.
  """

  # Make the filename lowercase
  filename = text.lower()

  # Remove characters that are not allowed in filenames
  allowed_chars = set(string.ascii_lowercase + string.ascii_uppercase + string.ascii_letters + string.digits + "-_. ")
  filename = ''.join([char for char in filename if char in allowed_chars])

  # Replace spaces with underscores
  filename = filename.replace(" ", "_")

  # Remove leading and trailing special characters
  filename = filename.strip("_-.")

  # If the filename is empty, use a placeholder
  if not filename:
    filename = fallback

  # Handle Unicode characters (optional)
#   filename = unicodedata.normalize('NFKD', filename).encode('ascii', 'ignore').decode('ascii')

  return filename

if __name__ == "__main__":
    parser = argparse.ArgumentParser(
        "Audio Dataset Append",
    )
    parser.add_argument("audio_file", type=str, help="Path to audio file")
    parser.add_argument("--entxt", type=str, help="Path to .en.txt transcript file", required=False)
    parser.add_argument("--ass", type=str, help="Path to .ASS file", required=False)
    parser.add_argument("--outdir", "-o", help="Output Directory", required=False, default=os.getcwd())

    args = parser.parse_args()

    if not args.entxt and not args.ass:
        raise Exception("Provide one of entxt or ass file input.")


    segments: list[tuple[float, float, str]] = [] # start time (seconds), end time (seconds), transcript

    ########## Read input into segments ##########
    if args.entxt:
        with open(args.entxt, "r") as f:
            en_messages = f.read()
        
        en_messages = [line.strip() for line in en_messages.split("\n") if line.strip() != ""]

        for msg in en_messages:
            _, transcript, _, start, end = break_line(en_messages)
            segments.append((start, end, transcript))
    elif args.ass:
        with open(args.ass, "r", encoding="utf-8-sig") as f:
            doc = ass.parse(f)
        
        for e in doc.events:
            if isinstance(e, ass.line.Dialogue) and e.style == '英':
                segments.append((e.start.total_seconds(), e.end.total_seconds(), e.text))
    
    ########## Segment audio and output to data/ ##########
    data_dir = os.path.join(args.outdir, "data")
    audio_basename = to_ascii_filename(os.path.splitext(os.path.basename(args.audio_file))[0])

    jsonl_entries = []

    original_audio = pydub.AudioSegment.from_file(args.audio_file)

    os.makedirs(os.path.join(data_dir, audio_basename))
    
    for seg in segments:
        start_ms = int(seg[0] * 1000.0)
        end_ms = int(seg[1] * 1000.0)

        segment_name = f"{audio_basename}-{start_ms}-{end_ms}.mp3"
        segment_path = os.path.join(data_dir, audio_basename, segment_name)
        segment_relpath = os.path.join("data", audio_basename, segment_name)

        segment = original_audio[start_ms:end_ms]

        segment.export(segment_path)

        jsonl_entries.append({"file_name": segment_relpath, "transcription": seg[2], "origin": audio_basename})
    
    ########## Append metadata.jsonl ##########
    with open("metadata.jsonl", "a") as f:
        for entry in jsonl_entries:
            f.write(json.dumps(entry) + "\n")