WSC-Eval-ASR/process_textgrid.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

import re
from pathlib import Path

def extract_first_item_text(textgrid_file):
    """提取第一个item的文本（直接拼接，不留空格）"""
    with open(textgrid_file, 'r', encoding='utf-8') as f:
        content = f.read()
    
    # 提取item [1]的全部内容
    first_item_match = re.search(r'item\s*\[1\]:.*?intervals\s*:\s*size\s*=\s*\d+(.*?)(?=item\s*\[2\]|\Z)', 
                                content, re.DOTALL)
    if not first_item_match:
        return ""
    
    # 提取所有非空text字段
    texts = re.findall(r'text\s*=\s*"(.*?)"', first_item_match.group(1))
    return "".join([t for t in texts if t.strip()])  # 直接拼接，不留空格

def process_textgrids(textgrid_dir, wav_dir, output_dir):
    """处理目录下所有TextGrid文件"""
    textgrid_dir = Path(textgrid_dir)
    wav_dir = Path(wav_dir)
    output_dir = Path(output_dir)
    
    output_dir.mkdir(parents=True, exist_ok=True)
    
    with open(output_dir/'wav.scp', 'w', encoding='utf-8') as f_wav, \
         open(output_dir/'text', 'w', encoding='utf-8') as f_text:
        
        for textgrid_file in textgrid_dir.glob('*.TextGrid'):
            audio_id = textgrid_file.stem
            transcript = extract_first_item_text(textgrid_file)
            
            # 写入wav.scp
            wav_path = wav_dir/f"{audio_id}.wav"
            f_wav.write(f"{audio_id} {wav_path}\n")
            
            # 写入text（直接拼接的文本）
            f_text.write(f"{audio_id} {transcript}\n")

if __name__ == "__main__":
    textgrid_dir = ""
    wav_dir = ""
    output_dir = ""
    
    process_textgrids(textgrid_dir, wav_dir, output_dir)
    print(f"处理完成！输出目录: {output_dir}")