In the previous sharing, we introduced the bark model of TTS (Text-to-Speech), which turned out to be quite impressive. However, this model does not support training with our own voices. Therefore, today we will share a method through which we can train our ideal voice and use it to speak different languages and sing.
The code we will use is: SoftVC VITS Singing Voice Conversion Fork. You can find it at this link: https://github.com/voicepaw/so-vits-svc-fork
This model is a neural network that can convert one voice into another singing voice. We will continue to use Colab for training and generating voices.
The official colab notebook provided is located at: https://colab.research.google.com/github/34j/so-vits-svc-fork/blob/main/notebooks/so-vits-svc-fork-4.0.ipynb
You can copy this code to your Google Drive and run it step by step. It's worth noting that the training process takes a very, very long time—long enough for my flowers to wither while waiting.
Fortunately, I purchased colab pro+, which can continue to run in the background for over 24 hours even after I close the webpage (I feel like I'm advertising for this service...)
However, when you hear the results of the training, you'll feel that all the waiting was worthwhile. Let's listen to the versions by "I know everything" Trump and "Let me be clear" Obama～
Let's compare it with Guo Degang's original version:
Now listen to Trump's version:
Now listen to Obama's version:
Also sharing a resource, many people have trained a lot of sound models that can be used directly:
https://huggingface.co/QuickWick/Music-AI-Voices/tree/main https://huggingface.co/search/full-text?q=so-vits-svc-4.0-models
If you don't train and just use it, it's very simple; you can set up a GUI locally.
python -m venv venv
source venv/bin/activate #Create a virtual environment
pip3 install -U so-vits-svc-fork #Install
svcg #Launch GUI
After launching the GUI, upload the model and configuration files downloaded from the internet to the Model path and Config path on the left 👈, place the audio file you want to modify on the right 👉 in the Input audio path, select the Output audio path as the save location for the generated file, and click run.
