Detailed setup guide

Session-only by default

Cloud BYOK keys are used only for the current session by default. If you explicitly enable “Remember API keys on this PC”, VoxBridge Studio stores them with Windows Credential Manager. Keys should not be described or handled as plain settings, logs or config-file values.

If you already set GROQ_API_KEY, GEMINI_API_KEY or GOOGLE_API_KEY in the environment, the app can use those as a fallback.

Where should you start?

Start on the download page. The portable ZIP package should be extracted into a separate folder, and VoxBridge Studio should be started from that extracted folder.

• open the download page
• download the portable ZIP package
• extract it into a separate folder
• always start the app from the extracted folder

Speech recognition is also compute-heavy

• distil-large-v3: recommended base ASR, 16 GB RAM and 6-8 GB+ VRAM recommended; slower on CPU.
• large-v3-turbo: faster large ASR option, 16 GB RAM and 6-8 GB+ VRAM recommended.
• large-v3: quality mode, 16-32 GB RAM and 8 GB+ VRAM recommended; can be very slow on CPU.

For the first test, use an existing SRT translation or a 1-3 minute video with clear speech.

Install VoxBridge Studio

1. Open the download page.
2. Download the portable ZIP package.
3. Extract it into a separate, easy-to-find folder.
4. Start VoxBridge Studio from the extracted folder.
5. Let the first-run setup assistant appear after launch.

Quick check

ffmpeg -version

If this prints a version number, FFmpeg is visible to the system.

If it is not found, either it is not installed or it is not in PATH. Manual path entry inside the app is the quickest fallback.

Step by step in PowerShell

# 1) Create the folder
New-Item -ItemType Directory -Force -Path C:\Tools\ffmpeg

# 2) Extract the FFmpeg package here
# Skip this if you already extracted it

# 3) Check whether ffmpeg.exe exists
Get-ChildItem C:\Tools\ffmpeg -Recurse -Filter ffmpeg.exe

# 4) Add the bin folder to the current session PATH
$env:Path += ";C:\Tools\ffmpeg\bin"

# 5) Verify it
ffmpeg -version
where.exe ffmpeg

The same in cmd

mkdir C:\Tools\ffmpeg
where ffmpeg
set PATH=%PATH%;C:\Tools\ffmpeg\bin
ffmpeg -version

If ffmpeg -version works only in the current terminal, the PATH change is temporary. In that case either set PATH permanently in Windows, or point the app directly to ffmpeg.exe.

Quick check

ollama --version
ollama list

These are manual troubleshooting commands. In normal use, the app check is enough.

Step by step in PowerShell

# 1) Check the version
ollama --version

# 2) See whether it responds
ollama list

# 3) Start the local service manually if needed
ollama serve

The same in cmd

ollama --version
ollama list
ollama serve

ollama serve is usually only needed if Ollama is not already running. If ollama list works right after install, you often do not need to start it separately.

Download the translation model

The setup window lets you choose a translation model and start the download in one place. For the first run, one model is enough.

1. Choose a model in the Translation model section of first-run setup.
2. If the machine is weaker or you are unsure, start with translategemma:4b.
3. If you have 24 GB RAM or 12 GB+ VRAM, translategemma:12b is a good base model.
4. Click Download selected model.
5. Wait until the download is complete.
6. Run the check again; model status should be ready.

Recommended models

ollama pull translategemma:4b
ollama pull translategemma:12b
ollama pull translategemma:27b

translategemma:4b is the weak-PC first-test and compatibility model, not the quality baseline. translategemma:12b is the recommended base model, but it is not a good first choice below 16 GB RAM. translategemma:27b usually gives better translation quality, but it needs strong hardware, should show a warning before use, and is not recommended below 32 GB RAM.

You do not need all three models. Pick one for the first test, and try another later if needed.

PowerShell / cmd commands

# Faster model for weaker machines
ollama pull translategemma:4b

# Balanced default model
ollama pull translategemma:12b

# Larger, usually better-quality model
ollama pull translategemma:27b

# Verify
ollama list

Quick runtime test

ollama run translategemma:4b "Translate to Hungarian: Hello, this is a short test."

ollama run translategemma:12b "Translate to Hungarian: Hello, this is a short test."

ollama run translategemma:27b "Translate to Hungarian: Hello, this is a short test."

If ollama run returns a sensible translation, the model is really running. The 27b model is listed separately because it is a heavy quality mode for strong machines.

4b, 12b or 27b?

• translategemma:4b: 3.3 GB, weak-PC first test or compatibility fallback; not the quality baseline
• translategemma:12b: 8.1 GB, recommended base model; start here with 24 GB RAM or 12 GB+ VRAM
• translategemma:27b: 17 GB, heavy quality mode with an explicit warning before use; 32-64 GB RAM or 24 GB+ VRAM recommended
• if you are unsure or the machine is weaker, start with 4b and a short test
• if quality matters more, try 27b only on strong hardware

What should be green?

• FFmpeg available
• Ollama running
• at least one translation model downloaded
• license or trial status ready
• the first test can start

On weaker machines, Ollama may respond slowly. A smaller model and another setup check are the best first steps.

First successful trial

Start with a short, non-private video, not with an important long course recording. A sidecar subtitle is the easiest first case.

1. Choose a 1-3 minute clip.
2. If a sidecar subtitle exists, start with that.
3. If not, try the audio-to-subtitle path.
4. Check the language-tagged output, for example VideoName.hu.srt.
5. Once that works, you are ready for longer material.

Support export

If it still does not start, create a support bundle and send it to support. It can include logs, settings and technical state information.

• email: support@voxbridgestudio.org
• short error description
• Windows version
• FFmpeg / Ollama / model status