# Brethof Voice Pro

> Commercial voice-to-text and offline translation software for Linux and Windows. Runs 100% offline on your own machine — transcribe 30 languages (with automatic recognition of 22 Chinese regional dialects) and translate across 38 — with a perpetual license. Local LoRA fine-tuning on your own voice is built in.

Brethof Voice Pro is a desktop dictation, transcription, and translation app built on the open-source Qwen3-ASR speech engine. Everything — inference, translation, noise reduction, fine-tuning — runs locally on CPU or an NVIDIA, AMD, or Intel GPU (Vulkan). No cloud, no API calls, no usage caps, no subscription. One purchase, perpetual license; optional annual update pass.

Current version: **v2.3.3**. Transcribes 30 languages (plus 22 Chinese dialects) at a measured 1.84% word error rate on the standard benchmark, and translates offline across 38 languages with Tencent Hunyuan MT2 (its 7B tier reaches 97.9% of Google Gemini 3.1 Pro on FLORES-200, XCOMET-XXL).

Key things the product does:
- Hotkey-anywhere dictation — press a key (default F9) or a mouse side button, speak, and text types into whatever app has focus
- Transcribe a saved audio/video file, a microphone session, or system audio (whatever is playing through your speakers)
- Offline translation across 38 languages — translate live transcriptions, or translate text and SRT/VTT subtitle files, with optional bilingual (source + translation) output, all on-device (Tencent Hunyuan MT2)
- Export plain text or timestamped SRT subtitles; an optional Forced Aligner add-on gives word-level timestamps
- Built-in DeepFilter noise suppression
- LoRA fine-tuning on your own voice corpus (collected automatically from your corrections) — local, NVIDIA GPU or CPU
- Hotwords, language pinning / auto-detect, and a correction loop that feeds the training set
- Runs as an MCP server (19 tools) so Claude Desktop and other AI agents can drive transcription and translation (paid license)

What is NOT yet available:
- macOS build — not supported, and no date is promised
- LoRA voice training on AMD / Intel GPUs — training runs on an NVIDIA GPU or CPU for now (transcription and translation already run on NVIDIA, AMD, and Intel via Vulkan)

## Product

- [Product Tour](https://brethof.ai/voice/tour/): Walkthrough of every screen and feature with screenshots
- [Features](https://brethof.ai/voice/features/): Full feature list in detail
- [Videos](https://brethof.ai/voice/videos/): Tutorial videos — Getting Started, Transcribe Popup, Custom Training, Tips & Tweaks, Release Demos
- [Download](https://brethof.ai/voice/download/): Linux and Windows binaries, v2.3.3
- [Pricing](https://brethof.ai/voice/pricing/): Free 14-day trial; perpetual Personal and per-seat Business licenses (region-adjusted pricing on the page)
- [Champions Program](https://brethof.ai/voice/champions/): Community-contributor recognition program — intake currently closed (may reopen)

## Support

- [Support & FAQ](https://brethof.ai/voice/support/): Troubleshooting, microphone setup, GPU requirements, hotkeys
- [ffmpeg setup](https://brethof.ai/voice/support/ffmpeg/): Platform-specific ffmpeg install steps for file transcription
- [Privacy Policy](https://brethof.ai/voice/privacy/)
- [Terms of Service](https://brethof.ai/voice/terms/)

## Updates & engineering

- [Updates feed](https://brethof.ai/voice/updates/): Chronological release notes and engineering posts
- [Why we moved from ONNX to GGUF](https://brethof.ai/voice/updates/gguf-engine/)
- [Why Brethof Voice Pro runs 100% offline](https://brethof.ai/voice/updates/why-offline/)
- [Launch announcement](https://brethof.ai/voice/updates/launch/)

## Account

- [Account portal](https://brethof.ai/voice/account/): License management, device activation, update pass
- [Register](https://brethof.ai/voice/account/register.html): Create an account

## Languages

Transcription covers 30 languages, with 22 Chinese regional dialects recognised automatically; offline translation covers 38 languages. Languages can be pinned for best short-phrase results or auto-detected. The site and app UI are localised into many languages, with the URL pattern `https://brethof.ai/voice/{lang}/...` and `<link rel="alternate" hreflang="{lang}">` set on every page.