Skip to main content

getAvailableModels()v4.0.518

Returns the timestamped Whisper models supported by transcribe().

models.ts
import {getAvailableModels} from '@remotion/whisper-webgpu';

const models = getAvailableModels();
console.log(models);

Return value​

Returns an array with the following properties:

name​

The model name accepted by APIs in this package.

modelId​

The upstream Hugging Face model identifier.

parameters​

The number of parameters in the model.

multilingual​

Whether the model supports languages other than English.

supportsTranslationv4.0.523​

Whether the model supports task: 'translate' in transcribe(). Translation converts speech into English. It is supported by multilingual non-turbo models only.

webGpuDownloadSize​

The estimated model weight download size in bytes. Browser caches and supporting model files may add a small amount.

Available models​

Available sizes are tiny, base, small, medium, and large-v3-turbo. English-only .en variants are available through medium.

The set of available models may change between versions as faster variants of these models are in development by Hugging Face. Call clearStaleModels() to remove cached models that are no longer available after upgrading.

Model hosting​

We noticed that hosting the models on R2 leads to faster downloading than through Hugging Face. Therefore, we mirrored the models to remotion.media, keeping them byte-identical.

Compatibility​

BrowsersServersEnvironments
Chrome
Firefox
Safari
Node.js
Bun
Serverless Functions

See also​