User Manual v5.0

S2S Portable — "Perfect Build" Portable Edition

Created by Hardy Brielbeck

⚠ IMPORTANT: Support Policy

This software is free for personal, non-commercial use (home use).

No official support is provided for personal use. Please use this manual to resolve any issues.

Commercial / professional use requires a valid license. Support is provided exclusively to licensed users.

Contact: Brielbeck@hotmail.de

1. About S2S Portable

S2S Portable is a bidirectional, real-time Speech-to-Speech translator that runs entirely offline on your Windows PC. It was built as a love letter to the VRChat community.

100% Portable — No installation. Unzip to a USB stick and run anywhere.
100% Private — All AI runs locally. Your voice never leaves your PC.
28 Languages — Full speech translation and a fully translated GUI.
Bidirectional — You speak, they hear the translation. They speak, you hear it back.
Swappable AI — Uses Faster Whisper for speech recognition, with interchangeable translation and TTS engines.

2. System Requirements

💡 Key Requirement: NVIDIA GPU

S2S Portable requires a dedicated NVIDIA graphics card for real-time performance. An RTX 3060 (12GB) or better is recommended. CPU-only mode is available but significantly slower.

OS: Windows 10 or Windows 11
GPU: NVIDIA RTX 3060 12GB or better (CUDA required)
RAM: 8GB minimum, 16GB recommended
Storage: ~5–10GB for models (fits on USB)
Audio: Microphone + speakers/headphones
Optional: VB-Audio Virtual Cable (for routing audio to apps)

GPU	VRAM	One-Way Latency	Rating
RTX 4090 / 5090	24GB	~0.5–1s	Ultra
RTX 4080	16GB	~0.7–1.3s	High
RTX 4070 Ti / 4070	12GB	~1–1.5s	High
RTX 3070	8GB	~1.5–2.5s	Medium
RTX 3060	12GB	~1.5–2.5s	Medium
RTX 3050	8GB	~2.5–4s	Low
CPU Only	N/A	~10–20s	Fallback

On a good PC (RTX 4070-class), expect about 4–5 seconds for a full bidirectional round-trip. High-end RTX cards achieve under 1 second one-way!

3. Quick Start Guide

Download — Get the portable zip from the download page.

Extract — Unzip the folder to your Desktop, USB stick, or any location.

Run — Double-click START.bat inside the folder. The app will initialize and open.

Select Languages — Choose your source language (what you speak) and target language (what they hear).

Speak! — Click "Start Translation" and begin talking. The AI will transcribe, translate, and speak the result.

4. Key Features

Bidirectional Translation

Full 2-way live translation. You speak in your language, your friend hears the translation in theirs — and vice versa. Perfect for real conversations.

Push-to-Talk (PTT)

Enable PTT so the app only listens when you hold a hotkey. Great for noisy environments.

Voice Mode & Text Mode

Switch between speaking into your mic or typing text to be translated and spoken aloud.

Chat Window

A chat-style interface showing the conversation flow with both your messages and translated responses.

Swappable AI Models

The architecture is modular. Faster Whisper handles speech recognition. Translation and TTS engines can be swapped for different models as better ones become available.

Three TTS Engines

Piper TTS — Fast, natural voices for European languages. 40+ languages, 100+ voices.
MeloTTS — High-quality emotional speech for Japanese, Korean, and Chinese.
pyttsx3 — CPU fallback using Windows system voices.

⚡ Qwen3 7B — Lightning-Fast Translation

NEW! Context-aware AI translation powered by Alibaba's Qwen3. Understands idioms, slang, and humor. Faster AND smarter than traditional translation on mid-to-high GPUs.

Setup Wizard

First launch walks you through language, target, and mode selection. No manual setup needed.

Mode Selector

Switch between Desktop/Calls mode (Zoom, Teams, Discord) and VR Gaming mode (VRChat, Resonite) with one click. Settings adjust automatically.

5. Supported Languages (28)

Every language below has full speech-to-speech translation support and a completely translated GUI:

English • Japanese • Korean • Chinese (Simplified) • Chinese (Traditional) • Chinese (Hong Kong) • Spanish • French • German • Italian • Portuguese • Portuguese (Brazil) • Russian • Ukrainian • Polish • Dutch • Swedish • Danish • Norwegian • Finnish • Czech • Slovak • Romanian • Hungarian • Greek • Turkish • Arabic • Vietnamese

6. Performance & GPU Tiers

The app automatically detects your hardware and recommends optimal settings. You can also manually select your GPU in the System tab.

⚡ Speed Tip

For the fastest possible setup: use Faster Whisper (Tiny) + your preferred translation model + Piper TTS + Turbo Audio Mode. On an RTX 4070 Ti or better, this achieves under 1 second one-way latency!

7. Using with VRChat, Discord, Zoom, Teams & WhatsApp

S2S Portable works with any voice-chat application through Virtual Audio Cable (VB-Cable) routing:

Install VB-Audio Virtual Cable (free download from vb-audio.com).
In S2S Portable, set the audio output to "CABLE Input (VB-Audio)".
In your voice app (Discord, VRChat, Zoom, etc.), set the microphone to "CABLE Output (VB-Audio)".
Now when S2S speaks the translation, it feeds directly into your voice app!

For bidirectional conversations (hearing their speech translated back to you), you may need a second VB-Cable. The app's Audio tab has a built-in Cable Test to verify routing.

For startup cable auto-detection, the post-wizard setup summary, Open Windows Sound, and a Quest / VRChat / Windows routing table, see the standalone Virtual Audio Cable setup guide.

8. Accessibility

Deaf Mode (Read Translations)

Opens a large, always-on-top overlay window showing all translations as big, easy-to-read text. Perfect for deaf or hard-of-hearing users.

Type-to-Speak (Write to Translate & Speak)

Type messages that get translated and spoken via TTS. Perfect for users who cannot speak but want to communicate in a different language.

9. Voice Cloning

Record 20 seconds of your voice, and OpenVoice V2 will adapt the translated speech to match your unique timbre. You'll sound like yourself, not a robot.

Go to the Voice Clone tab in Expert Mode.
Record or upload a 20-second audio clip of your voice.
Click "Process" — the AI creates your voice profile.
Enable voice cloning — all translations now use your voice!

10. Troubleshooting

App won't start

Make sure you extracted the full zip (don't run from inside the archive).
Run START.bat, not individual Python files.
Check that your antivirus isn't blocking the app.

No audio output

Check your audio output device in the Audio tab.
Make sure VB-Cable is installed if routing to another app.
Try the built-in Cable Test in the Audio tab.

Slow translation

Ensure you have an NVIDIA GPU with up-to-date drivers.
Use a smaller Whisper model (Tiny or Small) for faster STT.
Enable Turbo Audio Mode in the System tab.
Close other GPU-heavy applications (games, rendering, etc.).

Translation quality issues

Speak clearly and at a moderate pace.
Try switching between translation engines (NLLB vs Qwen3).
Short sentences translate better than long paragraphs.

Qwen3 / LLM not loading

Make sure a .gguf file is in the models/ folder.
Enable "Use Local LLM (Qwen)" in AI Models tab and restart.
If VRAM is full, use Whisper Tiny to free GPU memory.

⚠ Reminder: No Support for Personal Use

This software is provided "as-is" for personal use. No official support is given. Please use this manual to troubleshoot. Commercial users with a valid license receive dedicated support.

11. What's Automated

The following things are now fully automatic — no manual configuration needed:

Setup Wizard — First launch guides through language & mode selection.
Language & Voice Auto-Save — Changes saved to config.ini instantly.
Optimal VAD & ANTIGHOST — Best settings applied on every startup.
Smart Language Detection — Detects when source=target and passes through.
Auto TTS Voice Selection — Best voice picked when changing language.
Hardware Detection — GPU, VRAM, system scanned on startup.
Mode-Based Configuration — Switching modes adjusts all settings.

12. Credits & Donations

9+ months of intensive development combining open-source tools. Built as a love letter to the VRChat community by Hardy Brielbeck.

If you enjoy barrier-free conversations, consider supporting continued development:

PayPal: paypal.me/HardyBrielbeck