https://github.com/spycsh/quantized-bert-vits2

int8 quantization to Bert-VITS2
https://github.com/spycsh/quantized-bert-vits2

Last synced: 24 days ago
JSON representation

int8 quantization to Bert-VITS2

• Host: GitHub
• URL: https://github.com/spycsh/quantized-bert-vits2
• Owner: Spycsh
• License: apache-2.0
• Created: 2024-01-19T09:06:11.000Z (10 months ago)
• Default Branch: main
• Last Pushed: 2024-01-23T03:50:22.000Z (10 months ago)
• Last Synced: 2024-01-30T03:36:50.664Z (9 months ago)
• Language: Python
• Homepage:
• Size: 14.6 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        

This is some experiment code to support Bert-VITS2 int8 static quantization by using Intel Neural Compressor and Intel Extension For Transformers.

Bert-VITS2 is a great repo on multilanguage TTS, but it has not been optimized in inference speed or memory space. This repo target at the inference speed and memory space optimization. This repo is not an official project from Intel and should be experimental. Future features may be integrated into Intel Extension For Transformers NeuralChat.

```shell

git clone https://github.com/intel/intel-extension-for-transformers.git

cd intel-extension-for-transformers

git checkout spycsh/bvits-tts

cd ..

export PYTHONPATH=$(pwd)/intel-extension-for-transformers

python quantize.py --name=vits

python quantize.py --name=vits_bert_cn

python infer.py

# python reduce_nosie.py

```

For details, refer to this [PR](https://github.com/intel/intel-extension-for-transformers/pull/1139)