Local voice conditioning
Record or import a short, clean reference clip and keep the voice profile inside the local workspace.
Offline Windows studio · v1.0.0
Your voice workflow, on your machine.
XTTS Portable turns text and a consenting reference voice into WAV audio with a complete local runtime—no cloud synthesis, no account and no voice upload.
5.21 GiB Windows x64 ZIP64 · Non-commercial use only · Persian speech is experimental.
- Offline after setupModel and runtime included
- CUDA or CPUAutomatic fallback
- DE / FA interfaceGerman is the default UI
A complete local studio
From reference voice to WAV.
The portable package brings the interface, Python runtime, audio stack and XTTS-v2 model together in one Windows folder.
Practical file workflow
Type text or import common text, subtitle, markup and data formats, then create a standard WAV output.
Automatic acceleration
Use NVIDIA CUDA when available and fall back to CPU when the selected GPU path cannot run.
Bilingual interface
Start in German and switch to the Persian UI without rearranging the core workspace.
Sensitive by nature
Reference voices remain local.
A voice sample is biometric and personal data. XTTS Portable is built for on-device processing and does not require uploading your recording, text or generated audio to a synthesis service.
- Reference recordings stay in the local voices folder.
- Generated WAV files stay in your chosen local output path.
- No account, cloud inference or analytics is required.
- Only use a voice with the speaker's informed permission.
System requirements
A serious local model needs room.
XTTS Portable trades a large download and heavier local processing for privacy and independence from an online service.
Windows 10 or 11 x64
The package contains its own application runtime. Extract it to a normal writable folder.
16 GB RAM recommended
Keep at least 12 GB free for the extracted package, temporary work and generated audio.
NVIDIA GPU optional
A CUDA-capable GPU improves speed. CPU mode works but model loading and synthesis can be much slower.
Clean reference input
A microphone or a 6–20 second clean voice file gives the model a better reference.
Allow time to load
The local model is large. A cold start can take well over a minute on modest hardware.
Not required for synthesis
Once the complete package is extracted, everyday generation can remain offline.
Use responsibly
Consent first. Then create.
The technical ability to imitate a voice does not grant permission to use somebody's identity or to mislead an audience.
- Download and fully extractThe package is large; do not run the batch files from inside the archive.
- Run check_environment.batConfirm the runtime, model, audio devices and available CUDA or CPU path.
- Launch run.batChoose German or Persian, then record or import a clean reference voice.
- Generate and reviewListen critically before using the WAV file and keep outputs appropriately labelled.
XTTS Portable · v1.0.0
Build audio locally—with permission.
Download the complete Windows package for private, non-commercial voice work.