VoiceTransor

AI-Powered Speech-to-Text with Local Processing

VoiceTransor is a desktop application that converts audio to text using OpenAI's Whisper model, with optional AI text processing powered by Ollama. Everything runs locally on your computer - your data never leaves your machine.

中文版说明 (Chinese README) | Developer Guide

Features

• Accurate Speech Recognition - Powered by OpenAI Whisper
• GPU Acceleration - Automatic CUDA/MPS detection for faster processing
• AI Text Processing - Summarize, translate, or process transcripts with Ollama
• Multiple Export Formats - Save as TXT or PDF
• Privacy First - All processing happens locally
• Cross-Platform - Windows, macOS, and Linux support
• Multiple Languages - Supports 100 languages

Quick Start

1. System Requirements

Minimum:

• Windows 10 / macOS 10.15 / Linux
• 8GB RAM
• 5GB free disk space

Recommended for GPU Acceleration:

• NVIDIA GPU (GTX 900 series or newer)
• Driver version >= 525.60 (for CUDA support)

Note: GPU is optional - the app automatically detects your hardware and falls back to CPU if needed.

2. Installation

Windows:

1. Download VoiceTransor-v0.9.0-Windows-x64-Setup.exe from Releases
2. Run the installer and follow the setup wizard
3. Launch VoiceTransor from the Start Menu or Desktop shortcut

macOS / Linux:

• Coming soon - Currently only Windows installer is available
• For development setup, see Developer Guide

Important: You also need to install FFmpeg (see below).

3. Install FFmpeg (Required)

VoiceTransor needs FFmpeg for audio processing.

Windows:

• Download: https://www.gyan.dev/ffmpeg/builds/
• Choose "ffmpeg-release-essentials.zip"
• Extract and add to PATH (How?)

macOS:

brew install ffmpeg

Linux:

sudo apt install ffmpeg  # Ubuntu/Debian

4. First Use

1. Launch VoiceTransor
2. Click "Import Audio" (supports WAV, MP3, M4A, FLAC, etc.)
3. Click "Transcribe to Text"
4. Choose settings:
   • Model: base (recommended for most users)
   • Device: auto (automatically uses GPU if available)
   • Language: Auto-detect or select specific language
5. Wait for transcription (first time downloads the model ~140MB)
6. Export as TXT or process with AI

Optional: Install Ollama

Ollama enables AI-powered text processing (summarize, translate, etc.) - completely offline.

Installation:

1. Download from: https://ollama.com/download
2. Install and run ollama serve
3. Pull a model:

   ollama pull llama3.1:8b  # English
   ollama pull qwen2.5:7b   # Chinese/English

Note: Ollama works on both CPU and GPU - no special setup needed.

Performance

Transcription Speed (1 hour audio):

Hardware	Time
CPU (8-core)	~30-60 min
NVIDIA RTX 3060	~2-5 min
Apple M1 Pro	~3-6 min

GPU Compatibility:

• ✅ Single universal build - works on both GPU and CPU systems
• ✅ Automatic detection - no manual configuration needed
• ✅ NVIDIA GPUs (GTX 900+, RTX series) - uses CUDA acceleration
• ✅ Apple Silicon (M1/M2/M3) - uses Metal Performance Shaders
• ✅ Graceful fallback to CPU if GPU unavailable
• ℹ️ No separate CPU/GPU versions - one installer works for everyone

Documentation

• User Guide - Detailed usage instructions
• Installation Guide - Step-by-step installation
• Build Instructions - For developers
• 中文文档 - Chinese documentation

Troubleshooting

"ffprobe not found"

• Install FFmpeg and ensure it's in your PATH

Slow transcription

• Use smaller model (tiny or base)
• Check Device setting is auto or cuda
• Ensure GPU drivers are up to date

"Ollama is not running"

• Open terminal and run ollama serve

For more help, see USER_GUIDE.md

License

MIT License. See LICENSE for details.

Support

• GitHub Issues: https://github.com/leonshen/VoiceTransor/issues
• Email: voicetransor@gmail.com

---

Made with ❤️ using OpenAI Whisper and Ollama