install.ps1 script is NOT working - Unable to install!!!

[jdc4429]

(SoVits) C:\GPT-SoVITS>powershell -File install.ps1 -Device CU126 -Source HF
Active code page: 65001
[INFO]: Installing FFmpeg & CMake...

[ERROR] Command failed:
Message:
C:\Users\jeffc\anaconda3\Lib\site-packages\conda\base\context.py:892: FutureWarning: Adding the 'free' channel as it existed prior to conda 4.7. is deprecated and will be removed in 25.3. See https://docs.conda.io/projects/conda/en/stable/user-guide/configuration/free-channel.html for more details.
Command: $output = & conda install -y -q -c conda-forge @Args 2>&1
Location: C:\GPT-SoVITS\install.ps1:55
Call Stack:
at Invoke-Conda, C:\GPT-SoVITS\install.ps1: line 55
at , C:\GPT-SoVITS\install.ps1: line 140

(SoVits) C:\GPT-SoVITS>

[jdc4429]

I think I found the fix.. Replace the trap {} routine with this:

trap {
$msg = $.Exception.Message
if ($msg -match "FutureWarning|DeprecationWarning|free.*channel") {
continue
}
Write-ErrorLog $
}

Warnings are causing the script to exit!

[jdc4429]

PS - This would not be an issue if the windows release was actually up to date! Why is v4 not included???

[jdc4429]

Found a ton more issues with the script...

I fixed the install script and provide it here if anyone wants to use:

Param ( [Parameter(Mandatory=$true)][ValidateSet("CU124", "CU126", "CU128", "CPU")][string]$Device, [Parameter(Mandatory=$true)][ValidateSet("HF", "HF-Mirror", "ModelScope")][string]$Source, [switch]$DownloadUVR5 ) $global:ErrorActionPreference = 'Stop' function Write-ErrorLog { param ( [System.Management.Automation.ErrorRecord]$ErrorRecord ) Write-Host "n[ERROR] Command failed:" -ForegroundColor Red
if (-not $ErrorRecord.Exception.Message){
} else {
Write-Host "Message:" -ForegroundColor Red
$ErrorRecord.Exception.Message -split "n" | ForEach-Object { Write-Host " $_" } } Write-Host "Command:" -ForegroundColor Red -NoNewline Write-Host " $($ErrorRecord.InvocationInfo.Line)".Replace("r", "").Replace("n", "") Write-Host "Location:" -ForegroundColor Red -NoNewline Write-Host " $($ErrorRecord.InvocationInfo.ScriptName):$($ErrorRecord.InvocationInfo.ScriptLineNumber)" Write-Host "Call Stack:" -ForegroundColor DarkRed $ErrorRecord.ScriptStackTrace -split "n" | ForEach-Object {
Write-Host " $_" -ForegroundColor DarkRed
}

exit 1

}
trap {
$msg = $.Exception.Message
if ($msg -match "FutureWarning|DeprecationWarning|free.*channel") {
continue
}
Write-ErrorLog $
}
function Write-Info($msg) {
Write-Host "[INFO]:" -ForegroundColor Green -NoNewline
Write-Host " $msg"
}
function Write-Success($msg) {
Write-Host "[SUCCESS]:" -ForegroundColor Blue -NoNewline
Write-Host " $msg"
}
function Test-ZipValid {
param([string]$Path)
try {
Add-Type -AssemblyName System.IO.Compression.FileSystem
$zip = [System.IO.Compression.ZipFile]::OpenRead($Path)
$zip.Dispose()
return $true
} catch {
return $false
}
}
function Invoke-Conda {
param (
[Parameter(ValueFromRemainingArguments = $true)]
[string[]]$Args
)

& conda install -y -c conda-forge $Args

}
function Invoke-Pip {
param (
[Parameter(ValueFromRemainingArguments = $true)]
[string[]]$Args
)
$output = & pip install @Args 2>&1
$exitCode = $LASTEXITCODE
if ($exitCode -ne 0) {
$errorMessages = @()
Write-Host "Pip Install $Args Failed" -ForegroundColor Red
foreach ($item in $output) {
if ($item -is [System.Management.Automation.ErrorRecord]) {
$msg = $item.Exception.Message
Write-Host "$msg" -ForegroundColor Red
$errorMessages += $msg
}
else {
Write-Host $item
$errorMessages += $item
}
}
throw [System.Exception]::new(($errorMessages -join "`n"))
}
}
function Invoke-Download {
param (
[Parameter(Mandatory = $true)]
[string]$Uri,

    [Parameter()]
    [string]$OutFile
)
try {
    if ($OutFile -and (Test-Path $OutFile)) {
        $isZip = $OutFile -match "\.zip$"
        $isTar = $OutFile -match "\.tar\.gz$"
        if ($isZip -and -not (Test-ZipValid $OutFile)) {
            Write-Info "Removing corrupt/incomplete zip: $OutFile"
            Remove-Item $OutFile -Force
        } elseif ($isZip -and (Test-ZipValid $OutFile)) {
            Write-Info "$OutFile already exists and is valid, skipping download"
            return
        } elseif ($isTar) {
            Write-Info "Removing existing tar.gz to ensure clean download: $OutFile"
            Remove-Item $OutFile -Force
        }
    }
    $params = @{ Uri = $Uri }
    if ($OutFile) { $params["OutFile"] = $OutFile }
    $null = Invoke-WebRequest @params -ErrorAction Stop
} catch {
    Write-Host "Failed to download:" -ForegroundColor Red
    Write-Host "  $Uri"
    throw
}

}
function Invoke-Unzip {
param($ZipPath, $DestPath)
Expand-Archive -Path $ZipPath -DestinationPath $DestPath -Force
Remove-Item $ZipPath -Force
}
chcp 65001
if ($PSScriptRoot) { Set-Location $PSScriptRoot }
Write-Info "Installing FFmpeg & CMake..."
Invoke-Conda ffmpeg cmake
Write-Success "FFmpeg & CMake Installed"
$PretrainedURL = ""
$G2PWURL = ""
$UVR5URL = ""
$NLTKURL = ""
$OpenJTalkURL = ""
switch ($Source) {
"HF" {
Write-Info "Download Model From HuggingFace"
$PretrainedURL = "https://huggingface.co/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/pretrained_models.zip"
$G2PWURL = "https://huggingface.co/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/G2PWModel.zip"
$UVR5URL = "https://huggingface.co/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/uvr5_weights.zip"
$NLTKURL = "https://huggingface.co/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/nltk_data.zip"
$OpenJTalkURL = "https://huggingface.co/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/open_jtalk_dic_utf_8-1.11.tar.gz"
}
"HF-Mirror" {
Write-Info "Download Model From HuggingFace-Mirror"
$PretrainedURL = "https://hf-mirror.com/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/pretrained_models.zip"
$G2PWURL = "https://hf-mirror.com/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/G2PWModel.zip"
$UVR5URL = "https://hf-mirror.com/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/uvr5_weights.zip"
$NLTKURL = "https://hf-mirror.com/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/nltk_data.zip"
$OpenJTalkURL = "https://hf-mirror.com/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/open_jtalk_dic_utf_8-1.11.tar.gz"
}
"ModelScope" {
Write-Info "Download Model From ModelScope"
$PretrainedURL = "https://www.modelscope.cn/models/XXXXRT/GPT-SoVITS-Pretrained/resolve/master/pretrained_models.zip"
$G2PWURL = "https://www.modelscope.cn/models/XXXXRT/GPT-SoVITS-Pretrained/resolve/master/G2PWModel.zip"
$UVR5URL = "https://www.modelscope.cn/models/XXXXRT/GPT-SoVITS-Pretrained/resolve/master/uvr5_weights.zip"
$NLTKURL = "https://www.modelscope.cn/models/XXXXRT/GPT-SoVITS-Pretrained/resolve/master/nltk_data.zip"
$OpenJTalkURL = "https://www.modelscope.cn/models/XXXXRT/GPT-SoVITS-Pretrained/resolve/master/open_jtalk_dic_utf_8-1.11.tar.gz"
}
}
if (-not (Test-Path "GPT_SoVITS/pretrained_models/sv")) {
Write-Info "Downloading Pretrained Models..."
Invoke-Download -Uri $PretrainedURL -OutFile "pretrained_models.zip"
Invoke-Unzip "pretrained_models.zip" "GPT_SoVITS"
Write-Success "Pretrained Models Downloaded"
} else {
Write-Info "Pretrained Model Exists"
Write-Info "Skip Downloading Pretrained Models"
}
if (-not (Test-Path "GPT_SoVITS/text/G2PWModel")) {
Write-Info "Downloading G2PWModel..."
Invoke-Download -Uri $G2PWURL -OutFile "G2PWModel.zip"
Invoke-Unzip "G2PWModel.zip" "GPT_SoVITS/text"
Write-Success "G2PWModel Downloaded"
} else {
Write-Info "G2PWModel Exists"
Write-Info "Skip Downloading G2PWModel"
}
if ($DownloadUVR5) {
if (-not (Test-Path "tools/uvr5/uvr5_weights")) {
Write-Info "Downloading UVR5 Models..."
Invoke-Download -Uri $UVR5URL -OutFile "uvr5_weights.zip"
Invoke-Unzip "uvr5_weights.zip" "tools/uvr5"
Write-Success "UVR5 Models Downloaded"
} else {
Write-Info "UVR5 Models Exists"
Write-Info "Skip Downloading UVR5 Models"
}
}
switch ($Device) {
"CU124" {
Write-Info "Installing PyTorch For CUDA 12.4..."
Invoke-Pip torch --index-url "https://download.pytorch.org/whl/cu124"
}
"CU126" {
Write-Info "Installing PyTorch For CUDA 12.6..."
Invoke-Pip torch --index-url "https://download.pytorch.org/whl/cu126"
}
"CU128" {
Write-Info "Installing PyTorch For CUDA 12.8..."
Invoke-Pip torch --index-url "https://download.pytorch.org/whl/cu128"
}
"CPU" {
Write-Info "Installing PyTorch For CPU..."
Invoke-Pip torch --index-url "https://download.pytorch.org/whl/cpu"
}
}
Write-Success "PyTorch Installed"
Write-Info "Installing Python Dependencies From requirements.txt..."
Invoke-Pip -r extra-req.txt --no-deps
Invoke-Pip -r requirements.txt
Write-Success "Python Dependencies Installed"
$nltkTarget = (python -c "import sys; print(sys.prefix)").Trim()
if (-not (Test-Path "$nltkTarget\nltk_data")) {
Write-Info "Downloading NLTK Data..."
Invoke-Download -Uri $NLTKURL -OutFile "nltk_data.zip"
Invoke-Unzip "nltk_data.zip" $nltkTarget
Write-Success "NLTK Data Downloaded"
} else {
Write-Info "NLTK Data Exists"
Write-Info "Skip Downloading NLTK Data"
}
$jtalkTarget = (python -c "import os, pyopenjtalk; print(os.path.dirname(pyopenjtalk.file))").Trim()
if (-not (Test-Path "$jtalkTarget\open_jtalk_dic_utf_8-1.11")) {
Write-Info "Downloading Open JTalk Dict..."
Invoke-Download -Uri $OpenJTalkURL -OutFile "open_jtalk_dic_utf_8-1.11.tar.gz"
tar -xzf open_jtalk_dic_utf_8-1.11.tar.gz -C $jtalkTarget
Remove-Item "open_jtalk_dic_utf_8-1.11.tar.gz" -Force
Write-Success "Open JTalk Dict Downloaded"
} else {
Write-Info "Open JTalk Dict Exists"
Write-Info "Skip Downloading Open JTalk Dict"
}
Write-Success "Installation Completed"
`

[XXXXRT666]

I believe most people will not use the conda free channel in their configurations, but rather the default channel

[jdc4429]

I just wanted to get it to work. LOL (I never looked at the channel, not sure why it matters?)

I thought I was done after fixing the installer... But found a ton more errors when attempting to train.
Finally got it working.. Gloo was causing major issues so had to change for single GPU..

Needed to modify 3 files to get it working for V2ProPlus... s1_train.py, s2_train.py, bucket_sampler.py

Here is some more info for those trying to get it to work on Windows:

Install in Powershell Admin CMD promt.
Use powershell script from inside conda env - 'SoVITS': install-FINAL-FIXES.ps1

The script may fail with an error stating requirements.txt failed but in my case it did install correctly, I just reran the script again to get to completion message. Doing pip install requirements did not make any changes or give any errors so it actually didn't fail.

Installed with CU126

conda create -n SoVITS python=3.10
conda activate SoVITS
powershell -File install-FINAL-FIXES.ps1 -Device CU126 -Source HF (Can select from CU124, CU126, CU128)

To fix errors:
I am also getting this error:
ffmpeg.exe - Entry Point Not Found - The procedure entry point libintl_bind_textdomain_codeset could not be located in the dynamic link library C:\Users\jeffc\anaconda3\envs\SoVITS\Library\bin\gdk_pixbuf-2.0-0.dll.

This is a DLL version conflict — ffmpeg.exe is trying to load gdk_pixbuf-2.0-0.dll from your conda environment, but that DLL was compiled against a newer version of libglib that exports libintl_bind_textdomain_codeset differently. Here's how to fix it:

Fix 1: Downgrade libglib in your SoVITS environment (Most Reliable)
The root cause is that libglib=2.78.4=h55e6270_1 on Windows causes this symbol mismatch. The fix is to pin it to the *_0 build.

conda activate SoVITS
DIDN'T WORK - conda install libglib=2.78.4=_0 -c conda-forge
WORKS - conda install libglib=2.78.4=_0 gdk-pixbuf -c conda-forge --force-reinstall

Fix for:

File "C:\Users\jeffc\anaconda3\envs\SoVITS\lib\site-packages\torchaudio_torchcodec.py", line 84, in load_with_torchcodec
raise ImportError(
ImportError: TorchCodec is required for load_with_torchcodec. Please install torchcodec to use this function.

conda activate SoVITS
pip install torchcodec

Fix for:

-- Process 0 terminated with the following error:
Traceback (most recent call last):
File "C:\Users\jeffc\anaconda3\envs\SoVITS\lib\site-packages\torch\multiprocessing\spawn.py", line 87, in _wrap
fn(i, *args)
File "C:\GPT-SoVITS\GPT_SoVITS\s2_train.py", line 80, in run
dist.init_process_group(
File "C:\Users\jeffc\anaconda3\envs\SoVITS\lib\site-packages\torch\distributed\c10d_logger.py", line 83, in wrapper
return func(*args, **kwargs)
File "C:\Users\jeffc\anaconda3\envs\SoVITS\lib\site-packages\torch\distributed\c10d_logger.py", line 97, in wrapper
func_return = func(*args, **kwargs)
File "C:\Users\jeffc\anaconda3\envs\SoVITS\lib\site-packages\torch\distributed\distributed_c10d.py", line 1813, in init_process_group
default_pg, _ = _new_process_group_helper(
File "C:\Users\jeffc\anaconda3\envs\SoVITS\lib\site-packages\torch\distributed\distributed_c10d.py", line 2043, in _new_process_group_helper
backend_class = ProcessGroupGloo(
RuntimeError: makeDeviceForHostname(): unsupported gloo device

Removed gloo use in s2_train.py - Only allows for single GPU use now! - Use FIXED version
Removed gloo use in s1_train.py - Only allows for single GPU use now! - Use FIXED version
Fixed bucket_sampler.py - Use FIXED version

If anyone wants the changes for the FIXED files, leave a message with an email address and I will send if I see the request.

[jdc4429]

After a simple test, Version 2 Pro Plus is much better than version 3. Trained with only a 20 second clip with default training parameters and it sounds really close. I'm sure it will be even better with 5-10 minutes of training audio data.
Much better than v1, v2, and v3. Have not tried v4 yet as sources states v2 Pro Plus is the best.