免费在线视频转文字
使用AI驱动的转录将视频转为文字。上传视频文件、通过麦克风录音或粘贴URL。支持100多种语言、10多种模型，准确率超过98%。
1. 上传视频
上传MP4、MKV、MOV、WebM或AVI。音频会自动提取。
2. AI转录视频
AI提取并转录音轨，包含说话人标签和时间戳。
3. 导出与字幕
下载SRT/VTT格式的字幕。或将转录导出为TXT、DOCX、PDF。
视频转文字模型
选择适合您需求的AI模型——或让我们为您选择最佳模型。
准备好将视频转为文字了吗？
免费开始 →常见问题
Upload your video file or paste a video URL. STT.ai extracts the audio track automatically — no separate demux step — runs it through your chosen AI model, and returns the transcript plus SRT/VTT subtitles.
MP4, MKV, MOV, WebM, AVI, and other common containers are all supported. You don't need to extract the audio yourself — upload the video as-is.
Yes. Export the transcript as SRT or VTT for upload to YouTube, Vimeo, or any player, and the burn-subtitles tool can hardcode captions directly onto the video. MKV and MP4 also support attaching soft-subtitle tracks without re-encoding.
Yes. STT.ai includes 600 free minutes per month — about ten hours of video. Paid plans starting at $5/month add larger files, longer videos, and private transcripts.
Accuracy depends on the audio track inside the video — higher-bitrate audio (256 kbps+) transcribes better than heavily compressed soundtracks. Our best models reach 93-95% on clean dialogue.
Files up to 2 GB are supported on every plan. Free users get up to one hour of video per file; paid plans extend that to 8+ hours. For huge raw camera files, compress to H.264/AAC or use a URL upload.
Yes. Paste a public video URL from any of 1,300+ supported platforms and STT.ai fetches the video and extracts its audio automatically. DRM-protected or private videos must be downloaded manually first.
Yes. Speaker diarization labels each voice (Speaker 1, Speaker 2, ...) and you rename them in the editor — useful for interviews, panels, and multi-host video.
Yes. 100+ languages with auto-detection. You can also translate the finished transcript or subtitles into other languages with the subtitle-translator tool for a wider audience.
Export to SRT or VTT for subtitles, plus TXT, DOCX, PDF, or JSON for articles, show notes, and archives. JSON keeps machine-readable timestamps and speaker labels.
Yes. Video and the extracted audio are processed and deleted by default, and Pro plans add client-side encryption so transcripts are unreadable without your key. Nothing is used for training without explicit opt-in.
Most videos finish in a few minutes; a one-hour video typically takes 3-5 minutes depending on the model and current GPU load. Long videos queue and email you when they're done.