faster-whisper-gpu

# 🎙️ Faster Whisper GPU

High-performance local speech-to-text transcription using [Faster Whisper](https://github.com/SYSTRAN/faster-whisper) with NVIDIA GPU acceleration.

## ✨ Features

- **🚀 GPU Accelerated**: Uses NVIDIA CUDA for blazing-fast transcription
- **🔒 100% Local**: No data leaves your machine. Complete privacy.
- **💰 Free Forever**: No API costs. Run unlimited transcriptions.
- **🌍 Multilingual**: Supports 99 languages with automatic detection
- **📁 Multiple Formats**: Input: MP3, WAV, FLAC, OGG, M4A. Output: TXT, SRT, JSON
- **🎯 Multiple Models**: From tiny (fast) to large-v3 (most accurate)
- **🎬 Subtitle Generation**: Create SRT files with word-level timestamps

## 📋 Requirements

### Hardware
- **NVIDIA GPU** with CUDA support (recommended: 4GB+ VRAM)
- Or CPU-only mode (slower but works on any machine)

### Software
- Python 3.8+
- NVIDIA drivers (for GPU support)
- CUDA Toolkit 11.8+ or 12.x

## 🚀 Quick Start

### Installation

```bash
# Install dependencies
pip install faster-whisper torch

# Verify GPU is available
python -c "import torch; print(f'CUDA available: {torch.cuda.is_available()}')"
```

### Basic Usage

```bash
# Transcribe an audio file (auto-detects GPU)
python transcribe.py audio.mp3

# Specify language explicitly
python transcribe.py audio.mp3 --language pt

# Output as SRT subtitles
python transcribe.py audio.mp3 --format srt --output subtitles.srt

# Use larger model for better accuracy
python transcribe.py audio.mp3 --model large-v3
```

## 🔧 Advanced Usage

### Command Line Options

```bash
python transcribe.py <audio_file> [options]

Options:
  --model {tiny,base,small,medium,large-v1,large-v2,large-v3}
                        Model size to use (default: base)
  --language LANG       Language code (e.g., 'pt', 'en', 'es'). Auto-detect if not specified.
  --format {txt,srt,json,vtt}
                        Output format (default: txt)
  --output FILE         Output file path (default: stdout)
  --device {cuda,cp...

# 🎙️ Faster Whisper GPU - OpenClaw Skill

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![Python 3.8+](https://img.shields.io/badge/python-3.8+-blue.svg)](https://www.python.org/downloads/)
[![CUDA](https://img.shields.io/badge/CUDA-11.8%2B-green.svg)](https://developer.nvidia.com/cuda-downloads)

> High-performance local speech-to-text transcription using Faster Whisper with NVIDIA GPU acceleration.

## ✨ Why This Skill?

- **🔒 Privacy First**: Your audio never leaves your machine
- **⚡ GPU Accelerated**: 10-20x faster than CPU transcription
- **💰 Zero API Costs**: Unlimited transcriptions, forever free
- **🌍 99 Languages**: Automatic language detection
- **🎯 Perfect for OpenClaw**: Seamless integration with your agent workflows

## 🚀 Quick Start

### 1. Install Dependencies

```bash
pip install faster-whisper torch
```

### 2. Verify GPU Support

```bash
python -c "import torch; print(f'CUDA available: {torch.cuda.is_available()}')"
```

### 3. Transcribe!

```bash
python transcribe.py audio.mp3
```

## 📖 Usage Examples

### Basic Transcription
```bash
python transcribe.py meeting.mp3
```

### Portuguese Audio to SRT Subtitles
```bash
python transcribe.py podcast.mp3 --language pt --format srt --output podcast.srt
```

### High-Accuracy Mode
```bash
python transcribe.py interview.mp3 --model large-v3 --vad_filter --word_timestamps
```

### Translate to English
```bash
python transcribe.py japanese.mp3 --task translate --format txt
```

## 🛠️ Requirements

### Hardware
- NVIDIA GPU with 4GB+ VRAM (recommended)
- Or CPU-only mode (slower)

### Software
- Python 3.8+
- NVIDIA Drivers
- CUDA Toolkit 11.8+ or 12.x

## 📊 Performance

| Model | VRAM | Speed (RTX 4090) | Accuracy |
|-------|------|------------------|----------|
| tiny | 1 GB | ~32x realtime | Basic |
| base | 1 GB | ~16x realtime | Good |
| small | 2 GB | ~6x realtime | Better |
| medium | 5 GB | ~2x realtime | Great |
| large-v3 | 10...