# What's RVC

Last update: July 17, 2025

---

# Introduction

---

• RVC (Retrieval-Based Voice Conversion) is an advanced AI voice cloning software, developed by the RVC-Project team. It's considered the best free & open-source one to date.

• It was designed for desktop, requiring great specs to run it effectively, specially GPU for training models.

• Though it can be executed through the cloud & be used in any device, in case you don't meet the previous requirement.

• RVC doesn't have any major quality improvements since 2023, since its original devs are focused on other projects, RVC is hard to optimize, and it has limitations like non speech sounds such as realistic laughing, screaming, etc. Though, there are commmunity driven Forks that try to experiment with it, mostly about adding new features and performance improvements.

---

# Forks

---

• A fork is a copy of a main GitHub Project. It aims to make a modified version of the project, with improvements, new features & modifications.

• RVC has quite a few forks made by the community, each one meeting different needs for the user, and with its pros & cons.

• These are the main ones, along with their cloud-based counterparts:

  • # Mainline (Original RVC)

    • # Mainline Kaggle

    • # Mainline Colab

    • # Ilaria RVC Zero (Inference only)

  # ‎

  • # Applio

    • # Applio Colab

    • # Applio Kaggle

    • # Applio Lightning Ai

  # ‎

  • # RVC-AI-Cover-Maker-UI

    • # RVC-AI-Cover-Maker-UI Colab

    • # RVC-AI-Cover-Maker-UI Kaggle

  ‎

---

# FAQ

# Frequently asked questions.

---

# ‎

• As explained before, it depends on your needs. It's best to try them yourself.
• For local users, Applio is a great starting point. For cloud users you can use either the Applio Colab or applio kaggle.

# ‎

The minimum specs vary depending if it's for training models or inference.

SPEC	MINIMUM REQUIREMENT
GPU	NVIDIA GTX 900 Series / AMD RX580 (Mac isn't supported)
RAM	6GB
Storage	30 GB

SPEC	SUGGESTED REQUIREMENT
GPU	NVIDIA RTX 20 Series or later / AMD Radeon RX 5xxx or later (Mac isn't supported)
VRAM	8GB
RAM	8GB
Storage	30 GB

SPEC	MINIMUM REQUIREMENT
RAM	6GB
Storage	6 GB

SPEC	SUGGESTED REQUIREMENT
GPU	NVIDIA RTX 20 Series or later / AMD Radeon RX 5xxx or later / Apple M3
RAM	6GB
Storage	6 GB

NOTES:

• For inference, the storage requirement varies depending on the fork. It can be around 5 to 9 GB
• If you don't meet the minimum requirements, it's more convenient to use RVC on the cloud.

# ‎

• You can, but it's going to be slower, as they don't have CUDA cores.
• So it's more convenient using RVC through the cloud.
• If you're willing to use a slower version you can go ahead and follow this guide on how to get zluda working with Applio Zluda.

# ‎

• The total time depends on a lot of factors, like dataset length, batch size, pretrains, specs, etc.

• A 10 min dataset with RMVPE may take around 1 to 2 hours.

# ‎

• Yes, on Macs of recent generations.
• But you can only do inference & it's a little unstable.

# ‎

• If you're using RVC locally, no (the only exception would be Applio TTS as it uses Microsoft's Edge TTS as a base).
• If you're using it through the cloud, then yes.

# ‎

• There isn't an official one, but there's an unofficial complex blog to understand how it works.

---

# ‎