ContextQMD
Back to Voice Changer

Kaggle Tutorial

Hina_Mod_Kaggle_Real_Time_Voice_Changer.ipynb

latest6.0 KB
Original Source

<a href="https://www.kaggle.com/code/hinabl/public-w-okada-voice-changer?scriptVersionId=151765879" target="_blank"></a>

w-okada's Voice Changer | Kaggle

⬇ VERY IMPORTANT ⬇

You can use the following settings for better results:

If you're using a index: f0: RMVPE_ONNX | Chunk: 112 or higher | Extra: 8192

If you're not using a index: f0: RMVPE_ONNX | Chunk: 96 or higher | Extra: 16384

**Don't forget to select a GPU in the GPU field, <b>NEVER</b> use CPU!

Seems that PTH models performance better than ONNX for now, you can still try ONNX models and see if it satisfies you

You can always click here to check if these settings are up-to-date

Credits

Realtime Voice Changer by w-okada

Notebook files updated by rafacasari

Recommended settings by Raven

Modded again by Hina

Need help? AI Hub Discord » #help-realtime-vc

Kaggle Tutorial

Running this notebook can be a bit complicated.
After created your Kaggle account, you'll need to verify your phone number to be able to use Internet Connection and GPUs.
Follow the instructions on the image below.

<font color=blue>You can use <b>GPU P100</b> instead of GPU T4, some people are telling that <b>P100 is better</b>.</font>

Clone repository and install dependencies

This first step will download the latest version of Voice Changer and install the dependencies. It will take some time to complete.

python
# This will make that we're on the right folder before installing
%cd /kaggle/working/

!pip install colorama --quiet
from colorama import Fore, Style
import os

!mkdir Hmod
%cd Hmod
!git clone https://github.com/w-okada/voice-changer.git --depth=1 --quiet .
print(f"{Fore.GREEN}> Successfully cloned the repository!{Style.RESET_ALL}")
%cd server
!sed -i "s/-.-.-.-/Kaggle.Mod/" '../client/demo/dist/assets/gui_settings/version.txt'
!mv MMVCServerSIO.py Hmod.py
!sed -i "s/MMVCServerSIO/Hmod/" Hmod.py

print(f"{Fore.CYAN}> Installing libportaudio2...{Style.RESET_ALL}")
!apt-get -y install libportaudio2 -qq

print(f"{Fore.CYAN}> Installing pre-dependencies...{Style.RESET_ALL}")
# Install dependencies that are missing from requirements.txt and pyngrok
!pip install faiss-gpu fairseq pyngrok --quiet 
!pip install pyworld --no-build-isolation
print(f"{Fore.CYAN}> Installing dependencies from requirements.txt...{Style.RESET_ALL}")
!pip install -r requirements.txt --quiet

# Download the default settings ^-^
if not os.path.exists("/kaggle/working/Hmod/server/stored_setting.json"):
    !wget -q https://gist.githubusercontent.com/Rafacasari/d820d945497a01112e1a9ba331cbad4f/raw/8e0a426c22688b05dd9c541648bceab27e422dd6/kaggle_setting.json -O /kaggle/working/24apuiBokE3TjZwc6tuqqv39SwP_2LRouVj3M9oZZCbzgntuG /server/stored_setting.json
print(f"{Fore.GREEN}> Successfully installed all packages!{Style.RESET_ALL}")

print(f"{Fore.GREEN}> You can safely ignore the dependency conflict errors, it's a error from Kaggle and don't interfer on Voice Changer!{Style.RESET_ALL}")

Start Server using ngrok

This cell will start the server, the first time that you run it will download the models, so it can take a while (~1-2 minutes)

You'll need a ngrok account, but <font color=green>it's free</font> and easy to create!

1 - Create a free account at ngrok
2 - If you didn't logged in with Google or Github, you will need to verify your e-mail!
3 - Click this link to get your auth token, and replace YOUR_TOKEN_HERE with your token.
4 - (optional) Change to a region near to you

python
Token = 'Token_Here'
Region = "ap" # Read the instructions below

# You can change the region for a better latency, use only the abbreviation
# Choose between this options: 
# us -> United States (Ohio)
# ap -> Asia/Pacific (Singapore)
# au -> Australia (Sydney)
# eu -> Europe (Frankfurt)
# in -> India (Mumbai)
# jp -> Japan (Tokyo)
# sa -> South America (Sao Paulo)

# ---------------------------------
# DO NOT TOUCH ANYTHING DOWN BELOW!

%cd /kaggle/working/Hmod/server
    
from pyngrok import conf, ngrok
MyConfig = conf.PyngrokConfig()
MyConfig.auth_token = Token
MyConfig.region = Region
conf.get_default().authtoken = Token
conf.get_default().region = Region
conf.set_default(MyConfig);

import subprocess, threading, time, socket, urllib.request
PORT = 8000

from pyngrok import ngrok
ngrokConnection = ngrok.connect(PORT)
public_url = ngrokConnection.public_url

def wait_for_server():
    while True:
        time.sleep(0.5)
        sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
        result = sock.connect_ex(('127.0.0.1', PORT))
        if result == 0:
            break
        sock.close()
    print("--------- SERVER READY! ---------")
    print("Your server is available at:")
    print(public_url)
    print("---------------------------------")

threading.Thread(target=wait_for_server, daemon=True).start()

!python3 Hmod.py \
  -p {PORT} \
  --https False \
  --content_vec_500 pretrain/checkpoint_best_legacy_500.pt \
  --content_vec_500_onnx pretrain/content_vec_500.onnx \
  --content_vec_500_onnx_on true \
  --hubert_base pretrain/hubert_base.pt \
  --hubert_base_jp pretrain/rinna_hubert_base_jp.pt \
  --hubert_soft pretrain/hubert/hubert-soft-0d54a1f4.pt \
  --nsf_hifigan pretrain/nsf_hifigan/model \
  --crepe_onnx_full pretrain/crepe_onnx_full.onnx \
  --crepe_onnx_tiny pretrain/crepe_onnx_tiny.onnx \
  --rmvpe pretrain/rmvpe.pt \
  --model_dir model_dir \
  --samples samples.json

ngrok.disconnect(ngrokConnection.public_url)