> ## Documentation Index
> Fetch the complete documentation index at: https://docs.varg.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Whisper Transcription

> Audio-to-text transcription with OpenAI Whisper — fal and Groq variants

## Overview

Transcribe audio to text using OpenAI's Whisper model, available through two providers with different speed/cost profiles.

| Model ID                      | Provider | Speed    | Credits | \~Cost |
| ----------------------------- | -------- | -------- | ------- | ------ |
| `whisper`                     | fal      | Standard | 10      | \$0.10 |
| `whisper-large-v3`            | fal      | Standard | 10      | \$0.10 |
| `groq-whisper`                | Groq     | Fast     | 5       | \$0.05 |
| `groq-whisper-large-v3`       | Groq     | Fast     | 5       | \$0.05 |
| `groq-whisper-large-v3-turbo` | Groq     | Fastest  | 3       | \$0.03 |

## Quick start

<CodeGroup>
  ```typescript SDK theme={null}
  import { createVarg } from "vargai/ai"

  const varg = createVarg({ apiKey: process.env.VARG_API_KEY! })

  const result = await varg.transcriptionModel("whisper").generate({
    file: "https://example.com/audio.mp3",
  })

  console.log(result.text)
  ```

  ```bash cURL theme={null}
  curl -X POST https://api.varg.ai/v1/transcription \
    -H "Authorization: Bearer $VARG_API_KEY" \
    -H "Content-Type: application/json" \
    -d '{
      "model": "whisper",
      "file": "https://example.com/audio.mp3"
    }'
  ```
</CodeGroup>

## Parameters

<ResponseField name="file" type="string" required>
  URL or local path to the audio file. Supports mp3, wav, m4a, ogg, flac, webm.
</ResponseField>

<ResponseField name="language" type="string">
  Language code (e.g., `"en"`, `"es"`, `"fr"`). Auto-detected if not specified.
</ResponseField>

<ResponseField name="prompt" type="string">
  Optional context to guide transcription. Useful for domain-specific terms or names.
</ResponseField>

<ResponseField name="temperature" type="number" default="0">
  Sampling temperature. 0 = deterministic, higher = more creative (not usually needed for transcription).
</ResponseField>

## Choosing a model

| Scenario     | Recommended                               | Why                          |
| ------------ | ----------------------------------------- | ---------------------------- |
| Cheapest     | `groq-whisper-large-v3-turbo` (3 credits) | 70% cheaper than fal Whisper |
| Best quality | `whisper-large-v3` (10 credits)           | Full Large V3 model on fal   |
| Fast + cheap | `groq-whisper` (5 credits)                | Good balance                 |
| Default      | `whisper` (10 credits)                    | Reliable, well-tested        |

## Use with captions

Transcription is commonly used to generate captions for videos:

```tsx theme={null}
const narration = Speech({
  model: varg.speechModel("eleven_v3"),
  text: "Welcome to the future of video creation.",
})

<Clip duration={5}>
  <Video model={varg.videoModel("kling-v3")} prompt="futuristic city" duration={5} />
  {narration}
  <Captions source={narration} />
</Clip>
```

## Pricing

| Model                                    | Credits | USD    |
| ---------------------------------------- | ------- | ------ |
| `whisper` / `whisper-large-v3`           | 10      | \$0.10 |
| `groq-whisper` / `groq-whisper-large-v3` | 5       | \$0.05 |
| `groq-whisper-large-v3-turbo`            | 3       | \$0.03 |

## Related models

<CardGroup cols={2}>
  <Card title="ElevenLabs Speech" icon="microphone" href="/models/speech/elevenlabs">
    Generate speech from text (the reverse operation).
  </Card>
</CardGroup>