Audio & Video to Subtitles (AI)

Transcribe speech into SRT, VTT or text with Whisper, in your browser

Runs in your browser

Drop an audio or video file

MP3, WAV, M4A, OGG, MP4, WebM and anything else your browser can play

The file stays on your device. Only the AI model is downloaded, once, and kept by your browser.

The first run downloads the model. Later runs start right away. Long files take a while on computers without WebGPU.

How does AI transcription in the browser work?

This tool runs OpenAI's open Whisper speech-recognition model on your own computer with transformers.js. Your browser decodes the audio track of the file, resamples it to 16 kHz, and the model turns each 30-second window into text with start and end times. Those timed segments become SRT or WebVTT cues, or a plain transcript. The model is downloaded once from the Hugging Face Hub and cached by the browser; the audio itself is never sent anywhere. It uses your graphics card through WebGPU when available, and the processor otherwise.

Use cases

  • Add subtitles to a video for YouTube, TikTok, CapCut or Premiere without typing them
  • Get a text transcript of an interview, lecture, meeting or podcast
  • Turn a voice message or a short audio note into text
  • Translate speech in another language into English subtitles

When to use?

Use it for clear speech in files up to about an hour. The Fast model suits short clips and quick drafts; Balanced is a good default; Accurate is noticeably better on accents and noisy audio but is a larger download and slower without a graphics card. Choose the spoken language when you know it, since auto-detect can guess wrong on short clips. Always read the result before publishing: names and technical words are the usual mistakes, and the Subtitle Converter can shift the timing afterwards.

Examples

  1. Make SRT subtitles for a video

    Input

    A 2-minute MP4 interview in Vietnamese. Model: Balanced. Spoken language: Vietnamese. Output: Same language.

    Output

    An .srt file with about 30 timed segments in Vietnamese, ready to load in a player or editor.

  2. Translate a talk into English subtitles

    Input

    A Spanish podcast clip (M4A). Model: Accurate. Output: Translate to English.

    Output

    English subtitles with the original timing, as SRT, VTT or plain text.

Questions and answers

Is my audio uploaded to a server?

No. The file is decoded and transcribed in your browser. The only download is the AI model itself, from the Hugging Face Hub, which your browser then caches.

Which model should I choose?

Start with Balanced. Use Fast for short clips or slower computers, and Accurate for accents, background noise or important work, if you can wait for the larger download.

Why is it slow?

Speech recognition is heavy work. With WebGPU (recent Chrome, Edge and Safari) it runs on the graphics card and is much faster; otherwise it runs on the processor. Keep the tab open until it finishes.

Which languages are supported?

Whisper recognizes about 100 languages. Accuracy is best for widely spoken ones such as English, Spanish, Chinese, Japanese and Vietnamese. Translation always goes into English.

Sources and standards

All Tools