alexturkey/voxcpm2-turkce-ses-tasarimi

What it solves

This project provides a local, privacy-focused way to design and use synthetic voices on modest hardware, specifically targeting a 4 GB GPU. It solves the problem of voice drift—where long-form audio loses its consistency—by using a reference-based, sentence-by-sentence generation method that keeps the voice timbre and style stable.

How it works

It uses the VoxCPM2 model paired with a patched audio engine. You provide a text description of the voice you want, and the system generates several variants. Once you pick a favorite, the system uses that specific recording as a reference to read your text. By processing text one sentence at a time using a fixed seed and reference, it ensures the output remains consistent from start to finish. It includes a local web interface and command-line tools for both designing new voices and cloning your own (with explicit permission).

Who it’s for

It is designed for users who want to generate high-quality synthetic speech locally without relying on cloud services. It is particularly useful for creators who need consistent voice narration for videos, audiobooks, or podcasts, and it is optimized for users with limited VRAM (like a 4 GB GTX 1050 Ti) or Apple Silicon Macs.

Highlights

• Runs fully locally with no data sent over the network.
• Includes 14 ready-made synthetic voices in 8 languages (Turkish, English, French, German, Spanish, Russian, Chinese, and Japanese).
• Optimized for low-memory hardware (peak VRAM ~2.8 GB).
• Includes a one-command installer for Windows and support for Apple Silicon.
• Provides a "Voice cloning" feature for personal use with consent-based safeguards.
• Allows for precise control over pronunciation via a custom dictionary.