2. V2EX
  3. [分享创造] 做了一个纯本地离线运行的语音转文字工具： Free Local Transcription

[分享创造] 做了一个纯本地离线运行的语音转文字工具： Free Local Transcription

mikicat 2026-08-31 21:00 1

大家好，


平时经常有一些包含敏感信息或者不想上传到第三方服务器的音频（如会议记录、采访、个人备忘等）需要转文字。市面上很多工具本质上都是套壳调用云端 API ，既有数据泄露风险，长期用也有不小的 API 成本。


结合自己的模型优化经验，我做了一个完全基于本地算力运行的音频转文字工具：Free Local Transcription


🔗 体验地址： https://freelocaltranscription.com/


主要特点
100% 本地运行 / 零数据外传：推理完全在设备本地完成，音频文件不会上传到任何服务器，断网状态下也能保证隐私与安全。


无需 API Key / 完全免费：依靠本地算力运行，不需要填 API Key ，也没有按分钟计费的困扰。


即开即用：界面保持极简，拖入音频即可直接进行本地转写。


目前的权衡与折衷
相比直接走云端高性能 GPU 集群，本地推理的速度会直接取决于用户的设备配置与硬件性能，老旧设备上可能存在内存占用较高的情况。


目前正在持续优化本地推理效率和交互体验，非常欢迎大家体验并拍砖：


在不同硬件配置下的运行速度与占用表现如何？


还有哪些日常使用中的刚需功能（如导出格式、时间戳、长音频切片等）值得优先加入？


感谢大家的时间与反馈！

最新回复 (6)
  • bdphin 08-31 21:26
    1楼
    在网页版上上传了一个两分钟的音频，选择了慢速模式，同时选择了挪威语，结果完全不能识别。

    识别结果：
    I NG NG NG NGJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJ
  • bdphin 08-31 21:29
    2楼
    又试了一个 3 分钟的音频，慢速模式，同时选择英语，也是不能识别。

    识别结果：
    I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. I'm going to learn. hat that that that that that that that that that that that that that that that that that that that that
  • WessonC 09-01 08:47
    3楼
    我也做过，
    用 qwen ASR ，时间戳+切片+speaker 识别
  • iyuanyi 09-01 08:55
    4楼
    拥挤的赛道 https://freeaudiototext.com/zh
  • jackOff 09-01 09:21
    5楼
    funasr 这个开源项目感觉完胜啊
  • aero99 09-01 09:29
    6楼
    试了很多，最后还是选择了 whisper
* 帖子来源V2EX
附近帖子
返回
飞读
随机推荐