Name: Whisper GPU Speech-to-Text API | Fast, Accurate Audio Transc
Brand: quickpod
SKU: api-quickpod-aiquick-whisper
Availability: InStock
Author: API Market

Anil Singh

# Whisper GPU — Speech-to-Text API

## Overview
The Whisper GPU — Speech-to-Text API provides high-quality audio transcription using the `faster-whisper large-v3` model. This powerful API efficiently converts uploaded audio files into text, complete with detailed word-level timing data and optional language hints. Leveraging an RTX 5060 Ti GPU, it ensures fast and accurate transcription for a wide range of applications. Developers can send an audio file and receive a comprehensive transcript, including details essential for playback synchronization, content review, or advanced downstream analysis.

## Features
*   **High Accuracy Transcription:** Utilizes the `faster-whisper large-v3` model for superior speech-to-text accuracy.
*   **Word-Level Timestamps:** Provides precise start and end times for each word, enabling perfect synchronization with audio playback.
*   **Segmented Output:** Transcripts are broken down into logical segments, each with its own timing and average log probability.
*   **Language Auto-Detection:** Automatically detects the spoken language if no hint is provided.
*   **Multi-Language Support:** Supports transcription in 99 languages.
*   **Translation Mode:** Directly translates spoken audio into English text, regardless of the source language (`task=translate`).
*   **Wide Audio Format Support:** Accepts various audio formats supported by ffmpeg, including MP3, WAV, M4A, OGG, FLAC, and WebM.
*   **Large File Support:** Processes audio files up to 200 MB in size.
*   **Detailed Response:** Includes the full concatenated transcript, detected language, language probability, and audio duration.
*   **GPU Accelerated:** Runs on an RTX 5060 Ti GPU for enhanced performance.

## Use Cases
*   **Meeting Transcription & Summarization:** Automatically convert spoken dialogue from meetings into searchable text, facilitating easy review, minute-taking, and information retrieval for business and corporate environments.
*   **Podcast Indexing & Search:** Generate accurate transcripts for podcast episodes, making their content fully searchable and improving discoverability for listeners and content creators in the media industry.
*   **Caption Generation for Video Content:** Create precise captions and subtitles for videos, enhancing accessibility for hearing-impaired audiences and improving SEO for platforms in the entertainment and e-learning sectors.
*   **Voice Assistant & Command Processing:** Power voice-driven applications by transcribing spoken commands into text, enabling seamless interaction and control for smart devices and software in the consumer electronics and IoT industries.
*   **Clinical Documentation & Medical Transcription:** Assist healthcare professionals by transcribing doctor-patient consultations or dictations into text, streamlining medical record-keeping and reducing manual data entry for the healthcare industry.



## API Documentation

### Base URL
All API requests should be made to: `https://prod.api.market/api/v1/quickpod/aiquick-whisper`

### Authentication
Authentication is handled via API key. Include your API key in the request header:
```
x-api-market-key: YOUR_API_MARKET_KEY
```


### Endpoints

#### `POST /v1/transcribe`
Transcribe an audio file to text with word-level timestamps.

**Description:**
Accepts any audio format supported by ffmpeg (MP3, WAV, M4A, OGG, FLAC, WebM). Set `task=translate` to output English regardless of the source language. Input file is deleted immediately after inference completes.

**Request:**
`POST https://prod.api.market/api/v1/quickpod/aiquick-whisper/v1/transcribe`
`Content-Type: multipart/form-data`

**Headers:**
*   `x-api-market-key`: `YOUR_API_MARKET_KEY` (Required)

**Body (multipart/form-data):**

| Parameter | Type   | Description                                                                     | Default     | Example  |
| :-------- | :----- | :------------------------------------------------------------------------------ | :---------- | :------- |
| `file`    | binary | Audio file (max 200 MB). **Required.**                                          |             |          |
| `language`| string | ISO 639-1 language code hint (e.g., `en`, `es`, `fr`). Auto-detected if omitted. | Auto-detect | `en`     |
| `task`    | string | Operation to perform. `transcribe` converts speech to text. `translate` converts speech to English text. | `transcribe`| `translate`|

**Example Request (cURL):**
```bash
curl -X POST "https://prod.api.market/api/v1/quickpod/aiquick-whisper/v1/transcribe" \
  -H "x-api-market-key: YOUR_API_MARKET_KEY" \
  -F "file=@/path/to/your/audio.mp3" \
  -F "language=en" \
  -F "task=transcribe"
```

**Response (200 OK):**
```json
{
  "text": "Hello, this is a test recording.",
  "language": "en",
  "language_probability": 0.9998,
  "duration_seconds": 3.84,
  "segments": [
    {
      "id": 0,
      "start": 0,
      "end": 3.84,
      "text": "Hello, this is a test recording.",
      "avg_logprob": -0.21
    }
  ],
  "words": [
    {
      "word": "Hello",
      "start": 0,
      "end": 0.52,
      "prob": 0.9985
    },
    {
      "word": ",",
      "start": 0.52,
      "end": 0.52,
      "prob": 0.9985
    },
    {
      "word": "this",
      "start": 0.74,
      "end": 0.94,
      "prob": 0.9845
    },
    {
      "word": "is",
      "start": 0.94,
      "end": 1.09,
      "prob": 0.9972
    },
    {
      "word": "a",
      "start": 1.09,
      "end": 1.15,
      "prob": 0.9983
    },
    {
      "word": "test",
      "start": 1.4,
      "end": 1.64,
      "prob": 0.9974
    },
    {
      "word": "recording.",
      "start": 1.64,
      "end": 2.24,
      "prob": 0.9981
    }
  ]
}
```

**Possible Error Responses:**
*   `413 File exceeds 200 MB limit`
*   `422 Unsupported format or processing error`

#### `GET /health`
Service health check.

**Request:**
`GET https://prod.api.market/api/v1/quickpod/aiquick-whisper/health`

**Headers:**
*   `x-api-market-key`: `YOUR_API_MARKET_KEY` (Required)

**Example Request (cURL):**
```bash
curl -X GET "https://prod.api.market/api/v1/quickpod/aiquick-whisper/health" \
  -H "x-api-market-key: YOUR_API_MARKET_KEY"
```

**Response (200 OK):**
```json
{
  "ok": true,
  "service": "whisper-api",
  "model": "large-v3",
  "device": "cuda"
}
```

#### `GET /v1/models`
List loaded model and available sizes.

**Request:**
`GET https://prod.api.market/api/v1/quickpod/aiquick-whisper/v1/models`

**Headers:**
*   `x-api-market-key`: `YOUR_API_MARKET_KEY` (Required)

**Example Request (cURL):**
```bash
curl -X GET "https://prod.api.market/api/v1/quickpod/aiquick-whisper/v1/models" \
  -H "x-api-market-key: YOUR_API_MARKET_KEY"
```

**Response (200 OK):**
```json
{
  "loaded": "large-v3",
  "device": "cuda",
  "compute_type": "float16",
  "available": [
    "tiny",
    "base",
    "small",
    "medium",
    "large-v2",
    "large-v3"
  ]
}
```

## Status Codes
*   **200 OK**: The request was successful, and the response body contains the requested data or the result of the operation.
*   **201 Created**: (Not applicable for this API's current endpoints, but generally indicates a resource was successfully created.)
*   **400 Bad Request**: The server cannot process the request due to malformed syntax, invalid request parameters, or other client-side errors. Check your request body and headers.
*   **401 Unauthorized**: The request lacks valid authentication credentials (e.g., missing or invalid `x-api-market-key`). Ensure your API key is correct and present.
*   **403 Forbidden**: Your API key is valid, but you do not have sufficient permissions to access the requested resource or perform the action.
*   **404 Not Found**: The requested API endpoint or resource does not exist. Verify the URL path.
*   **413 Payload Too Large**: The audio file sent for transcription exceeds the maximum allowed size of 200 MB.
*   **422 Unprocessable Content**: The request was well-formed, but the server was unable to process the input, for example, due to an unsupported audio format or a processing error during transcription.
*   **429 Too Many Requests**: You have sent too many requests in a given amount of time. Implement exponential backoff for retries.
*   **500 Internal Server Error**: A generic error message indicating an unexpected condition was encountered on the server. This usually points to a problem on the API provider's side.
*   **503 Service Unavailable**: The server is currently unable to handle the request due to temporary overload or maintenance. Retry the request after a short delay.

## Rate Limits
Specific rate limits are applied to ensure fair usage and service stability. If you exceed the allocated rate limit, you will receive a `429 Too Many Requests` HTTP status code. It is recommended to implement an exponential backoff strategy when retrying requests after encountering a 429 error to avoid exacerbating the issue. For higher throughput needs, please contact support.

## Error Handling
Effective error handling is crucial for building robust applications. This API uses standard HTTP status codes to indicate the outcome of a request, accompanied by JSON error messages for detailed information.

**How to Interpret Error Responses:**
Always check the HTTP status code of the response. For codes other than `2xx`, expect an error payload in JSON format, providing more context about the issue.

**Common Errors and How to Resolve Them:**
*   **400 Bad Request:**
    *   **Resolution:** Review the API documentation for the specific endpoint you are calling. Ensure all required parameters are included, that their types are correct (e.g., binary for file, string for language code), and that the request body is correctly formatted (e.g., `multipart/form-data` for file uploads).
*   **401 Unauthorized:**
    *   **Resolution:** Verify that your `x-api-market-key` header is present in your request and that the key is correct and active.
*   **413 Payload Too Large:**
    *   **Resolution:** Ensure your audio file does not exceed the 200 MB limit. Consider compressing your audio or splitting it into smaller segments if necessary.
*   **422 Unprocessable Content:**
    *   **Resolution:** Check the format of your audio file. Ensure it's one of the supported types (MP3, WAV, M4A, OGG, FLAC, WebM) and that the file is not corrupted.

**Recommended Retry Strategies for Transient Errors:**
For transient errors like `429 Too Many Requests` or `503 Service Unavailable`, implement a retry mechanism with exponential backoff.
1.  **Wait and Retry:** After receiving a transient error, wait for a short period before retrying the request.
2.  **Exponential Backoff:** Increase the waiting time exponentially after each failed retry. For example, wait 1 second, then 2 seconds, then 4 seconds, and so on.
3.  **Maximum Retries:** Set a reasonable limit for the number of retries (e.g., 3-5 times) to prevent indefinite loops.
4.  **Jitter:** Add a small random delay to the backoff period to prevent all clients from retrying simultaneously, which can further overload the server.

**Error Response Example:**
```json
{
  "error": {
    "code": "invalid_parameter",
    "message": "The audio file is too large. Maximum size is 200 MB.",
    "status": 413
  }
}
```





## FAQ

**Q: What is the maximum audio file size supported?**

A: The API supports audio files up to 200 MB.

**Q: What audio formats are supported?**

A: We support any audio format that `ffmpeg` can handle, including MP3, WAV, M4A, OGG, FLAC, and WebM.

**Q: Can I transcribe audio in languages other than English?**

A: Yes, the `faster-whisper large-v3` model supports 99 languages. You can specify the `language` parameter (ISO 639-1 code) to hint at the language, or it will be auto-detected.

**Q: What is the difference between `transcribe` and `translate` tasks?**

A: The `transcribe` task converts speech into text in the detected or specified original language. The `translate` task converts speech into English text, regardless of the original language spoken in the audio.

**Q: Does the API provide word-level timestamps?**

A: Yes, the transcription response includes detailed `words` objects, each with a `start` and `end` timestamp, and a `prob` (probability) for the word.

**Q: Is my audio data secure?**

A: Input files are deleted immediately after the inference process completes, ensuring your data privacy.

**Q: How can I improve transcription accuracy?**

A: Providing high-quality audio with minimal background noise and clear speech will yield the best results. Additionally, specifying the `language` parameter can improve accuracy if you know the spoken language.

## Contact Us
If you have any questions or need assistance, you can reach out to us through the embedded chat on the bottom right corner of the page or send an email to hello@magicapi.com.

Get high-quality, GPU-accelerated speech-to-text with Whisper GPU. Supports 99 languages, word-level timing, and translation. Powered by faster-whisper large-v3

Whisper GPU Speech-to-Text API | Fast, Accurate Audio Transc

Free