Skip to content

🎥 Make videos programmatically with React

Make web transcriber WebGPU-only

Jonny Burger26d9a9ebbc7d
diff --git a/packages/convert/app/components/transcribe/modelSelector.tsx b/packages/convert/app/components/transcribe/modelSelector.tsx
index 77cdcf8c7c..e78c9bc601 100644
--- a/packages/convert/app/components/transcribe/modelSelector.tsx
+++ b/packages/convert/app/components/transcribe/modelSelector.tsx
@@ -44,8 +44,7 @@ export default function ModelSelector({
<div className="text-left">
<div className="font-medium">{model.name}</div>
<div className="text-xs text-muted-foreground">
- {formatBytes(model.webGpuDownloadSize)} WebGPU ·{' '}
- {formatBytes(model.wasmDownloadSize)} WASM
+ {formatBytes(model.webGpuDownloadSize)} WebGPU
</div>
</div>
</SelectItem>
diff --git a/packages/convert/app/components/transcribe/state.ts b/packages/convert/app/components/transcribe/state.ts
index 3d32395aaf..ee0ea31c56 100644
--- a/packages/convert/app/components/transcribe/state.ts
+++ b/packages/convert/app/components/transcribe/state.ts
@@ -1,8 +1,5 @@
import type {Caption} from '@remotion/captions';
-import type {
- ResolvedWhisperWebGpuBackend,
- WhisperWebGpuModelLoadProgress,
-} from '@remotion/whisper-webgpu';
+import type {WhisperWebGpuModelLoadProgress} from '@remotion/whisper-webgpu';
export type TranscriptionState =
| {
@@ -17,11 +14,9 @@ export type TranscriptionState =
}
| {
type: 'transcribing';
- backend: ResolvedWhisperWebGpuBackend;
}
| {
type: 'done';
result: Caption[];
- backend: ResolvedWhisperWebGpuBackend;
}
| {type: 'error'; message: string};
diff --git a/packages/convert/app/components/transcribe/transcribeAudio.tsx b/packages/convert/app/components/transcribe/transcribeAudio.tsx
index e12946a923..567903cc81 100644
--- a/packages/convert/app/components/transcribe/transcribeAudio.tsx
+++ b/packages/convert/app/components/transcribe/transcribeAudio.tsx
@@ -43,7 +43,8 @@ export default function TranscribeAudio({
file: await sourceToBlob(source),
});
- const {backend} = await loadWhisperModel({
+ await loadWhisperModel({
+ backend: 'webgpu',
model: selectedModel,
onProgress: (progress) =>
setState(() => ({
@@ -54,11 +55,10 @@ export default function TranscribeAudio({
setState(() => ({
type: 'transcribing',
- backend,
}));
const transcription = await transcribe({
- backend,
+ backend: 'webgpu',
channelWaveform: waveform,
model: selectedModel,
});
@@ -66,7 +66,6 @@ export default function TranscribeAudio({
setState(() => ({
type: 'done',
result: toCaptions({whisperWebGpuOutput: transcription}).captions,
- backend,
}));
} catch (error) {
setState({
@@ -122,7 +121,7 @@ export default function TranscribeAudio({
<strong className="font-brand">Transcribing {name}</strong>
</div>
<div className="tabular-nums text-muted-foreground font-brand text-sm">
- <span>Using {state.backend.toUpperCase()}</span>
+ <span>Using WebGPU</span>
</div>
</div>
</>
@@ -131,7 +130,7 @@ export default function TranscribeAudio({
{state.type === 'done' ? (
<>
<Card className="p-3 text-sm text-muted-foreground">
- Transcribed with {state.backend.toUpperCase()}
+ Transcribed with WebGPU
{state.result.length === 0 ? ' · No speech detected' : null}
</Card>
<div className="h-4" />
diff --git a/packages/convert/test/transcribe-webgpu.test.tsx b/packages/convert/test/transcribe-webgpu.test.tsx
new file mode 100644
index 0000000000..d784863059
--- /dev/null
+++ b/packages/convert/test/transcribe-webgpu.test.tsx
@@ -0,0 +1,19 @@
+import {afterEach, expect, test} from 'bun:test';
+import {cleanup, render} from '@testing-library/react';
+import ModelSelector from '../app/components/transcribe/modelSelector';
+
+afterEach(() => cleanup());
+
+test('offers WebGPU model downloads without a WASM alternative', () => {
+ const rendered = render(
+ <ModelSelector
+ selectedModel="tiny.en"
+ setSelectedModel={() => undefined}
+ disabled={false}
+ />,
+ );
+
+ const model = rendered.getByRole('combobox', {name: 'Whisper model'});
+ expect(model.textContent).toContain('119.7 MB WebGPU');
+ expect(model.textContent).not.toContain('WASM');
+});