README.md

metadata

license: mit
language:
  - ar
  - cs
  - de
  - el
  - en
  - id
  - hi
  - fa
  - fr
  - ne
  - nl
  - 'no'
  - sw
  - sr
  - zh
  - vi
  - tr
  - uk
  - ru
  - ro
  - pt
  - pl
  - nl
  - hu
  - es
  - cs
pipeline_tag: text-to-speech

Guidelines for Converting Piper ONNX Model

References:

• https://github.com/nihui/ncnn-android-piper
• https://github.com/OHF-Voice/piper1-gpl
• https://huggingface.co/datasets/rhasspy/piper-checkpoints

Steps to convert Piper checkpoints to NCNN models:

1. Checkout the correct version of the piper repository:

   git clone [https://github.com/OHF-Voice/piper1-gpl](https://github.com/OHF-Voice/piper1-gpl)
   cd piper1-gpl
   git checkout 113931937cf235fc8all1afd1ca4be209bc6919bc7
   

2. Apply the necessary patch:

   # Ensure 'piper1-gpl.patch' is available
   git apply piper1-gpl.patch
   

3. Set up the Python environment and install dependencies:

   python3 -m venv .venv
   source .venv/bin/activate
   python3 -m pip install -e .[train]
   

4. Download a Piper checkpoint file (.ckpt) from Hugging Face: https://huggingface.co/datasets/rhasspy/piper-checkpoints

5. Install the PNNX model converter:

   pip install -U pnnx
   

6. Obtain the export_ncnn.py script.

7. Run the conversion script on your checkpoint file:

   # Replace with your actual file
   python export_ncnn.py (language code).ckpt (e.g., en.ckpt, fa.ckpt, ...)