PA Inference is an Android app that provides a UI for audio inference. By project gpt-sovits-onnx-rs, inferring on mobile device is made possible. By providing trained models and reference audios, audio will be inferred using the sounds of trained model.
With its tri-lingual design, text-to-speech conversion can be easily completed by just providing the text to be converted, then clicking the language to infer.
- Converted GPT-SoVITS v2 models in .onnx format as described in gpt-sovits-onnx-rs
- Model is available on Hugging Face (Private)
- 3-10s reference audio in .wav format (If Hugging Face model is used, reference audio is already bundled, and installation can be skipped)
- Download/compile and install the app
- Create a folder containing the following 3 folders
- yue
- zh
- en
- Put converted models and required models in the folders accordingly
- Put reference audios in the folders accordingly, naming it
ref.wav
- Select the root folder containing
yue,zh, anden - Modify the reference text which matches the ref.wav
- Enter the text to infer
- Press the language to infer
- Model will be loaded and audio will be inferred
- If same language is being inferred consecutively, model will only be loaded once to improve performance
Audio will be played and shown in Inference Results. By clicking ➦, audio can be shared. By clicking ✖, audio will be deleted.
- You are allowed to download the source code, compile and install on your own device.
- You are not allowed to redistribute any part of the code and claim that is your work.