melotts.axera

MeloTTS demo on Axera AX650

目前模型分成了encoder、flow、decoder三部分，encoder部分尚未转成axmodel（目前通过onnxruntime运行）。

TBD：

• encoder转成axmodel
• 效果与官方repo对齐

For Chinese input(Run on board)

locale-gen C.utf8
update-locale LANG=C.utf8

Python(Run on board)

Requirements

apt-get install libsndfile1-dev
mkdir /opt/site-packages
pip3 install -r requirements.txt --prefix=/opt/site-packages

Put
export PYTHONPATH=$PYTHONPATH:/opt/site-packages/local/lib/python3.10/dist-packages
export PATH=$PATH:/opt/site-packages/local/bin
to /root/.bashrc

Run

python3 onnx_infer_stream.py --sentence 要生成语音的句子

Cpp(Cross compile)

Download BSP

bash download_bsp.sh

Build

bash build.sh

Run on board

./build/install/bin/melotts -l ../models/melo_lexicon_zh.txt -t ../models/melo_tokens.txt -e ../models/enc-sim.onnx -f ../models/flow.axmodel -d ../models/decoder.axmodel --g ../models/g.bin -w test_cn.wav -s 爱芯元智半导体股份有限公司，致力于打造世界领先的人工智能感知与边缘计算芯片。服务智慧城市、智能驾 驶、机器人的海量普惠的应用