Guide for RVC's Realtime Voice Changer Client


This guide will be in the same style/format as Blancdot's w-okada guide for simplicity sake. Credits go to the legend. Thanks to Codename for additional input.


Last update: August 12th, 2024: voice models to try out added



Personal opinion + Recommendations + Overview


I would only recommend the RVC Realtime voice changer if you have a third party program like SteelSeries Sonar for noise gating and noise suppression, since the inbuilt settings on RVC for those are not great. If you have a headset mic, it will constantly pick up your headphone sounds of other people speaking if you don't use Sonar or other.


At this point in time, for all cases (Nvidia, AMD, Intel, Mac) I would recommend the modified w-okada version. Nvidia users may also set up the original w-okada environment manually by installing requirements instead if they have the knowledge on how to do that. Blanc_dot has a guide on Github.
Check out this guide for the modified w-okada version: https://rentry.co/ForkVoiceChangerGuide


Download


Download NVIDIA


NVIDIA

Latest as of 24th April 2024: RVCNvidiaLatest

If you have disk space worries, you can install the voice changer-only but it requires a manual installation. Check out C0dename's guide


Download AMD/INTEL GPUs


AMD & INTEL ARC

Latest as of 24th April 2024: RVCAMD_IntelLatest

If you want the latest version and/or have disk space worries, you can install the voice changer-only but it requires a manual installation. Check out C0dename's guide


Virtual Audio Cable


Virtual Cable is what you will need to use the voice changer on Discord & Games

  • Run setup64 after extracting the zip to a new folder
  • After installing the Virtual Cable, it changes your default audio system. Click on Yes when it asks you to open the audio device settings (or if you closed the window already, press WIN+R, type "mmsys.cpl"), and change your Recording and Playback devices back to your normal microphone and normal heapdhones. Make sure to do this for communications device aswell (right click sour normal mic or headphones -> set as default communication device)

Opening


  • Unzip your RVCLatest or RVC1006 that you downloaded to your Desktop or C:\ drive
  • Open the folder and depending on your GPU, you will open:
    • NVIDIA: go-realtime-gui or go-realtime-gui.bat
    • AMD/INTEL: go-realtime-gui-dml or go-realtime-gui-dml.bat

Once it's running in your CMD window DO NOT CLOSE IT, let it fully download everything, It will launch the voice changer after everything is downloaded


GUI settings explanation


IF YOU DON'T SEE A PICTURE HERE, PLEASE LET ME KNOW ON DISCORD "shadicti"


Load model

  • You upload your voice model on the first slot
  • You upload your index model on the second slot (if you do not plan on using index, PLEASE do change it anyway, because the default index file sometimes causes the program to crash)

Audio device

  • USE the SAME ending (MME) for Input and Output!

If you are an advanced user: the general thumbrule is ASIO > WASAPI > MME. How to enable Wasapi is explained in the Extras section.

On the voice changer, you do:

  • INPUT: Your microphone (MME)
  • OUTPUT: Line 1 Virtual Cable (MME)

On discord and games, you do:

  • INPUT: Line 1 Virtual Cable
  • OUTPUT: Your headphones

General settings

  • Response threshold: Noise gate so not everything gets picked up. However, I have experienced that it cuts off words and syllables, so I use something like SteelSeries Sonar or Voicemeeter to do my noise gating. I would keep it at -60 if you experience similar issues
  • Pitch settings: This is the tune/pitch. If you are a male and you are using a female voice, you want to aim between 10 to 14 and then go up and down depending on what you need. -12 if you are female to male and adjust, and male to male keep it at 0 and go up/down depending on if you think the model is too high/low pitch
  • Formant pitch shifting: manipulates the resonant peaks of the formants without affecting the pitch of the vocal track. This is a new feature - feel free to try it, but is not a must
  • Index Rate: Uses the index to force the accent of the voice model more. If this value is too high, you might experience autotune. Best to either use 0 or a low rate up to 0.5
  • Loudness factor: Volume of the voice changer output
  • pitch detection algorithm: Choose RMVPE or FCPE. FCPE is slightly faster so you can lower the delay, but if you experience any inconsistent pronounciations, then switch to RMVPE. I recommend testing both

Performance settings

  • Sample length: Delay. Underneath this section I will provide a guidance for different types of GPUs. It is unfinished, I would appreciate it if you help me with more information on GPUs that are not listed or might be inaccurate
  • Number of CPU processes used for harvest pitch algorithm: Ignore this, unless you are using "harvest" for whatever reason. Put it to any number
  • Fade length: Crossfade between inferenched chunks. High value will sound clearer, usually this is maxed out
  • Extra inference time: This controls the quality of the voice model (increases computional complexity), so if your GPU and/or CPU are not of newer gen, go with a lower number, but in almost all cases, you can max this out and have the best sounding voice model. On W-Okada, you use a very low number for this because the CPU usage cranks up to 100% instantly - but here on RVC realtime, you can max this out at no CPU cost
  • Input noise reduction: As it says, it adds noise reduction to your microphone before entering voice changer. I never use this personally
  • Output noise reduction: If your voice model is noisy, enable this. I personally always enable this no matter what
  • Phase Voncorder: can interpolate information present in the frequency and time domains of audio signals by using phase information extracted from a frequency transform. Just try it out, most times this makes no difference. I keep it on

GPU Chart for sample length + Extra (delay + quality)


for NVIDIA

Nvidia GPU's Sample length + Extra
RTX xx90 (for example 3090) 0.05 - 0.10 sample length + 2.70 extra
RTX xx80 Ti (for example 3080 Ti) 0.06 - 0.10 sample length + 2.70 extra
RTX xx80 (for example 3080) 0.08 - 0.12 sample length + 2.70 extra
RTX xx70 Ti (for example 3070 Ti) 0.10 - 0.14 sample length + 2.70 extra
RTX xx70 (for example 3070) 0.12 - 0.16 sample length + 2.70 extra
RTX xx60 Ti (for example 3060 Ti) 0.13 - 0.18 sample length + 2.70 extra
RTX xx60 (for example 3060) 0.15 - 0.20 sample length + 2.70 extra
RTX xx50 (for example 3050) 0.20 - 0.30 sample length + 2.70 extra
GTX 16-series 0.40 - 0.45 SL + 2.70 Extra
GTX 10-series no information

If it still lags

Raise the sample length until it doesn't. Limit fps or reduce ingame quality to give the voice changer more performance


for AMD

AMD GPU's Sample length + Extra
7xxx XT cards 0.13 - 0.20 SL + 2.70 Extra
6xxx XT cards 0.15 - 0.25 SL + 2.70 Extra
5xxx XT cards 0.25 - 0.40 SL + 2.70 Extra
RX 580 no information
RX 570 no information
RX 560 no information

If it still lags

Raise the sample length until it doesn't. Limit fps or reduce ingame quality to give the voice changer more performance


Extras


Reduce more delay


Reducing more delay guide

https://rentry.co/LessDelayWasapi


Voice Models to try out


You might need to connect your account to weights.gg to be able to download these models

Click on the 3 dots (...) on weights.gg models, then Download model.

Female:
Natural girl voice made by Mares
Duckus Egirl voice made by lusbert
Psych2Go voice made by Dan (this is another Duckus Egirl version)

Male:
Bob Ross voice made by dieseldog34
Markiplier voice made by hobqueer


Help


Common Errors


sounddevice.PortAudioError: Error opening Stream: Unanticipated host error [PaErrorCode -9999]: 'Failed to load ASIO driver' [ASIO error 0]

Even if it makes no sense, this is caused by the default index file, not even from ASIO itself. If you never changed the index, choose a random custom one - it does not even have to be the same as the voice model you are using. Just any random one will fix this.

sounddevice.PortAudioError: Error opening Stream: Invalid sample rate [PaErrorCode -9997]

  • The quick fix for this is to use (MME) as your input and output. Make sure you select it, e.g. "Microphone name (MME)"
  • If using Wasapi: You did not match the sample rate of your virtual cable to your microphone (this is explained in How to reduce more delay with Wasapi
  • If you did match the sample rate and but this happens on WASAPI & RVC realtime regardless, you should get FlexASIO. Click here to open up the guide

Crackle Fixes (Windows)


Open Task Manager > Details

Right click audiodg.exe and set priority to High

Right click audiodg.exe again > set affinity > uncheck everything except CPU 2 (only ✅ CPU 2, turn off the rest)

With a program called ProcessLasso you can automate this to always be active, since Windows resets these sometimes.


Discord Crackle Fixes


If the voice sounds fine in the app AND it sounds fine in games, but ONLY sounds weird on discord, then:

  • Turn off Echo Cancellation
  • Turn off Noise Suppression (sometimes causes issues, maybe not. Check for yourself)

Discord Server

Discord Server

Join the AI Hub discord server, ask in #help-rvc

Edit
Pub: 23 Apr 2024 23:08 UTC
Edit: 12 Aug 2024 21:28 UTC
Views: 6199