XTTS v2 (Coqui)
Multilingual voice cloning in 6 seconds.
Open Source 4–6 GB VRAM
- Min VRAM
- 4 GB
- GPU class
- Entry GPU
- Quant
- FP16
Actually FreeNo SignupOpen SourceWatermark-Free
RVC (Retrieval-based Voice Conversion)
The voice-changer that took over Discord.
Open Source 4–8 GB VRAMRuns locally
Actually FreeNo SignupOpen SourceWatermark-FreeHobbyist-Friendly
Visit RVC (Retrieval-based Voice Conversion)Updated 2025-10-12 · Direct link
Hardware requirements
Runs locally · Entry GPU (6–8 GB)
Min VRAM
4 GB
Rec. VRAM
8 GB
Min RAM
8 GB
Rec. RAM
16 GB
Disk
10 GB
GPU class
Entry GPU
11.7+Apple Silicon ✓CPU-CapableQuant: FP16
Inference on 4 GB; training comfortable at 8 GB. CPU works but slow.
Screenshot placeholder · RVC (Retrieval-based Voice Conversion)
What is RVC (Retrieval-based Voice Conversion)?
RVC takes an existing audio clip and replaces the speaker's voice with a trained target. Different problem than TTS: you need a source recording, but the result preserves all the prosody and emotion of the original take. The standard tool for voice covers, dubbing, and content creation.
Pros & cons
Pros
- Preserves singing, emotion, accent of the source take
- Train a new voice in minutes from 10+ minutes of clean audio
- Massive community model library on HuggingFace / Civitai
- Gradio web UI ships in-box
Cons
- Requires a source recording — not a from-scratch TTS
- Quality depends on clean training data; noisy inputs → robotic output
What's actually free?
MIT-licensed; web UI included.
✓ Actually FreeNo SignupOpen SourceWatermark-Free
Alternatives
Bark
Suno's expressive transformer-based TTS.
Open Source 8–12 GB VRAM
- Min VRAM
- 8 GB
- GPU class
- Entry GPU
- Quant
- FP16
Actually FreeNo SignupOpen SourceWatermark-Free