Skip to main content

@remotion/whisper-web

warning

Unstable API: This package is experimental for the moment. As we test it, we might make a few changes to the API and switch to a WebGPU-based backend in the future.

Similar to @remotion/install-whisper-cpp but for the browser. Allows you to transcribe audio locally in the browser, with the help of WASM.

Installation​

npx remotion add @remotion/whisper-web

Required configuration​

@remotion/whisper-web uses a WebAssembly (WASM) backend that requires SharedArrayBuffer support. To enable this functionality, you need to configure cross-origin isolation headers in your application. This is a security requirement for using SharedArrayBuffer in modern browsers.

See: Important Considerations

Vite​

To use @remotion/whisper-web with Vite, you need to make two important changes:

  1. Exclude the package from Vite's dependency optimization to prevent known issues
  2. Configure the required security headers for SharedArrayBuffer support
import {defineConfig} from 'vite';

export default defineConfig({
  optimizeDeps: {
    // turn off dependency optimization: https://github.com/vitejs/vite/issues/11672#issuecomment-1397855641
    exclude: ['@remotion/whisper-web'],
  },
  // required by SharedArrayBuffer
  server: {
    headers: {
      'Cross-Origin-Embedder-Policy': 'require-corp',
      'Cross-Origin-Opener-Policy': 'same-origin',
    },
  },
  // ...
});

Example usage​

import {transcribe, canUseWhisperWeb, resampleTo16Khz, downloadWhisperModel} from '@remotion/whisper-web';

const file = new File([], 'audio.wav');
const modelToUse = 'tiny.en';

const {supported, detailedReason} = await canUseWhisperWeb(modelToUse);
if (!supported) {
  throw new Error(`Whisper Web is not supported in this environment: ${detailedReason}`);
}

console.log('Downloading model...');
await downloadWhisperModel({
  model: modelToUse,
  onProgress: ({progress}) => console.log(`Downloading model (${Math.round(progress * 100)}%)...`),
});

console.log('Resampling audio...');
const channelWaveform = await resampleTo16Khz({
  file,
  onProgress: (p) => console.log(`Resampling audio (${Math.round(p * 100)}%)...`),
});

console.log('Transcribing...');
const {transcription} = await transcribe({
  channelWaveform,
  model: modelToUse,
  onProgress: (p) => console.log(`Transcribing (${Math.round(p * 100)}%)...`),
});

console.log(transcription.map((t) => t.text).join(' '));

Functions​

License​

MIT