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

# Transcribe

> Extract and transcribe audio from a video

# Transcribe Command

The `transcribe` command extracts audio from a video file or YouTube video and transcribes it using Whisper. This is useful for generating text content from videos or for accessibility purposes.

## Usage

```bash theme={null}
transfer-learning transcribe PATH [OPTIONS]
```

## Arguments

| Argument | Description                       |
| -------- | --------------------------------- |
| `PATH`   | Path to video file or YouTube URL |

## Options

| Option              | Description                                             | Default        |
| ------------------- | ------------------------------------------------------- | -------------- |
| `--output-dir TEXT` | Output directory for transcripts                        | `transcripts/` |
| `--model TEXT`      | Whisper model to use (tiny, base, small, medium, large) | base           |
| `--device TEXT`     | Device to use for transcription (cpu, cuda)             | cpu            |
| `--language TEXT`   | Language code (auto for auto-detection)                 | auto           |
| `--format TEXT`     | Output format (txt, srt, vtt, json)                     | txt            |
| `--help`            | Show help message and exit                              | -              |

## Examples

### Transcribe a local video file

```bash theme={null}
transfer-learning transcribe path/to/video.mp4
```

### Transcribe a YouTube video

```bash theme={null}
transfer-learning transcribe "https://www.youtube.com/watch?v=dQw4w9WgXcQ"
```

### Transcribe with a specific model

```bash theme={null}
transfer-learning transcribe path/to/video.mp4 --model medium
```

### Transcribe with GPU acceleration

```bash theme={null}
transfer-learning transcribe path/to/video.mp4 --device cuda
```

### Transcribe to a specific format

```bash theme={null}
transfer-learning transcribe path/to/video.mp4 --format srt
```

## Output

The command generates a transcript file in the specified format in the output directory:

```
output_dir/
└── video_name_transcript.txt
```

## Whisper Models

The command supports the following Whisper models:

| Model  | Size  | Memory Required | Relative Speed |
| ------ | ----- | --------------- | -------------- |
| tiny   | 39M   | \~1GB           | \~32x          |
| base   | 74M   | \~1GB           | \~16x          |
| small  | 244M  | \~2GB           | \~6x           |
| medium | 769M  | \~5GB           | \~2x           |
| large  | 1550M | \~10GB          | 1x             |

The larger models provide better accuracy but require more memory and processing time.

## Output Formats

The command supports the following output formats:

* `txt`: Plain text transcript
* `srt`: SubRip subtitle format
* `vtt`: WebVTT subtitle format
* `json`: JSON format with timestamps and confidence scores

## JSON Format Example

When using the `json` format, the output will look like this:

```json theme={null}
{
  "segments": [
    {
      "id": 0,
      "start": 0.0,
      "end": 5.0,
      "text": "Hello, welcome to this video.",
      "confidence": 0.95
    },
    {
      "id": 1,
      "start": 5.0,
      "end": 10.0,
      "text": "Today we're going to talk about...",
      "confidence": 0.92
    }
  ],
  "language": "en",
  "duration": 120.5,
  "word_count": 150,
  "processing_time": 45.2
}
```

## Related Commands

* [Process Video](/cli/process-video)
* [Process YouTube](/cli/process-youtube)
* [Analyze](/cli/analyze)