Abstract: A tutorial (experience sharing) on the usage and debugging of the RVC real-time voice changer developed by 花儿不哭 and the virtual audio mixing software Voicemeeter.

I. Voicemeeter Setup

1. System Sound Settings

Configure your Windows sound settings as follows:

Setting Device
Default output device Voicemeeter Input
Default communication output Voicemeeter Input
Default input device Voicemeeter Out B1
Default communication input Voicemeeter Out B1

Note: At this point, system audio (e.g., software audio) will be output to Voicemeeter Input. The input side will not detect audio input because Voicemeeter Out B1 is not a physical sound card.

2. Voicemeeter Configuration

Interface Overview

  • A = Hardware Output (monitoring)
  • B = Virtual Output (output to applications)

Channel Layout

Section Description
Left Area Audio input from physical devices (e.g., microphone)
Middle Area System and software audio input
Right Area Master output control - A for hardware monitoring, B for software output

Routing Guide

  1. For Self-Monitoring:

    • Check A1 in the corresponding input channel
    • Audio will be routed to A1 and output to headphones for monitoring

  2. For Application Output:

    • Check B1 in the corresponding input channel
    • Software will receive processed audio through the default input device (Voicemeeter Out B1)

The physical input and output devices must be set to MME format.
Using other formats will affect voice changing quality and increase latency.

II. RVC Voice Changer

1. System Requirements

Minimum Configuration:

  • GPU: NVIDIA GTX 1060 6GB or AMD RX580 8GB
  • CPU: Intel Core or AMD Ryzen (quad-core or above)
  • RAM: 8GB or above
  • OS: Windows 10 or Windows 11

Recommended Configuration:

  • GPU: RTX 3060 8GB
  • CPU: Intel 12th Gen Core i5 or Ryzen 7000 series or above
  • RAM: 32GB or above
  • OS: Windows 10 or Windows 11

Note: AMD and Intel graphics cards currently have higher latency. NVIDIA RTX 20 series or higher is recommended for optimal performance.

2. Running the Application

After downloading RVC, navigate to the installation directory:

1
2
3
4
RVC/
├── assets/weights/    # Voice models (.pth files)
├── logs/              # Model index files (.index files)
└── go-realtime-gui.bat

Run go-realtime-gui.bat to start the real-time voice changer.

3. Basic Software Settings

Configuration steps:

  1. Select the .pth model file and corresponding .index index file (optional)
  2. Set device type to MME
  3. Input device: Select your physical microphone
  4. Output device: Select Voicemeeter AUX Input (separate from system sound)
  5. Enable “Use device sample rate”

4. Parameter Tuning Guide

Basic Parameters

Parameter Description Recommended Values
Response Threshold Filters audio below threshold Higher = may swallow words
Pitch Voice pitch adjustment Male→Female: ~12, Female→Male: ~-12, Same gender: ~0
Gender Factor Voice thickness Positive = thinner, Negative = thicker
Index Rate Model similarity enhancement 0 if no matching index file; increase for better similarity
Loudness Factor Output volume level Higher = louder

Pitch Detection Algorithms

Available Algorithms:

  • pm: Lowest latency, lowest CPU. Good for singing, poor for speech in noisy environments
  • harvest: Best quality, higher CPU. Increase process count to reduce latency
  • crepe: Similar to harvest, GPU-intensive. Mostly superseded by rmvpe
  • rmvpe ⭐: Excellent default performance, faster than harvest/crepe, minimal GPU usage, highly recommended

Advanced Settings

  • Sample Length: Determines latency. Lower = less latency, but may cause stuttering
  • Fade Length: Affects naturalness vs. latency trade-off
  • Extra Inference Time: 1-3 seconds recommended. Higher = better word recognition, affects CUDA usage

CPU Process Count (Harvest algorithm only):

  • Affects sound quality; not recommended above 4
  • Determines max CPU threads for pitch algorithm
  • Lower values reduce gaming performance impact but increase inference latency
  • If inference latency meets requirements, use the lowest possible value

III. Audio Workflow

VoiceMeeter Banana ( Recommended ):

Retrieval-based-Voice-Conversion:
Official Sources:

Community:

  • RVC Voice Changer Group: 1026902887

References