PaddlePaddle / PaddleSpeech

Easy-to-use Speech Toolkit including Self-Supervised Learning model, SOTA/Streaming ASR with punctuation, Streaming TTS with text frontend, Speaker Verification System, End-to-End Speech Translation and Keyword Spotting. Won NAACL2022 Best Demo Award.
https://paddlespeech.readthedocs.io
Apache License 2.0
11.11k stars 1.85k forks source link

[TTS] VITS dygraph-to-static、Paddle2ONNX、Paddle-Lite #2826

Open yt605155624 opened 1 year ago

yt605155624 commented 1 year ago
yt605155624 commented 1 year ago

Paddle2ONNX 报错 image

check https://github.com/PaddlePaddle/Paddle2ONNX/issues/1043

yt605155624 commented 1 year ago

x86 image arm image

check https://github.com/PaddlePaddle/Paddle-Lite/issues/10045

PR:

目前二分定位到全图第一个 cumsum 的输入精度是不对的,也就是“出问题的地方是从这里到最开头的位置,大概率在子图” 6217f3a68951e02135754cbe57c23b5b