faster whisper从多媒体语音材料中抽取出文本-3

Avatar photoByAdil
"""
批量转录当前目录下的 .mp3 文件，使用 faster-whisper
"""

import os
import logging
import sys
from pathlib import Path
from faster_whisper import WhisperModel

logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger = logging.getLogger('Extrat Text')

# ================== 配置区 ==================
MODEL_SIZE = "small"          # 可选: tiny, base, small, medium, large
DEVICE = "cpu"                # cpu 或 cuda
COMPUTE_TYPE = "int8"         # int8, float16, float32 (CPU 推荐 int8)
VAD_FILTER = True             # 启用语音活动检测，去除静音
OUTPUT_FORMAT = "txt"         # 只输出 .txt
VERBOSE = True               # 是否显示详细日志
# ===========================================

def transcribe_audio(audio_path: Path, model: WhisperModel) -> str:
    """转录单个音频文件，返回文本内容"""
    print(f"转录: {audio_path.name} → {audio_path.stem}.txt")

    segments, info = model.transcribe(
        str(audio_path),
        language=None,           # 自动检测
        beam_size=5,
        vad_filter=VAD_FILTER,
        vad_parameters=dict(min_silence_duration_ms=500),
        word_timestamps=False,
    )

    text_lines = []
    for segment in segments:
        line = segment.text.strip()
        text_lines.append(line)
        if VERBOSE:
            logger.info(".", end="", flush=True)
            # print(f"[{segment.start:06.2f}s --> {segment.end:06.2f}s] {line}", flush=True)

    return "n".join(text_lines)


def main():
    print("=== faster-whisper 批量转录 ===")

    current_dir = Path(".")
    mp3_files = sorted(current_dir.glob("*.mp3"))

    if not mp3_files:
        print("未找到 .mp3 文件，退出。")
        return

    # 加载模型（只加载一次）
    print(f"正在加载模型 {MODEL_SIZE} ({DEVICE}, {COMPUTE_TYPE})...")
    model = WhisperModel(MODEL_SIZE, device=DEVICE, compute_type=COMPUTE_TYPE)

    processed = 0
    for mp3_path in mp3_files:
        txt_path = mp3_path.with_suffix(".txt")

        if txt_path.exists():
            print(f"跳过: {txt_path.name} 已存在")
            continue

        try:
            text = transcribe_audio(mp3_path, model)
            txt_path.write_text(text, encoding="utf-8")
            processed += 1
        except Exception as e:
            print(f"错误转录 {mp3_path.name}: {e}", file=sys.stderr)

    print(f"全部完成！共处理 {processed} 个文件。")


if __name__ == "__main__":
    main()

Similar Posts