Open PLL-L opened 8 months ago
我该如何改正,提升我的生成速度呢,谢谢
我找到了解决方法了,我尝试将frontend_en.py中的preprocess_english 模块直接调用,速度提升了许多
就是我想再问一下,这个模型的生成英文速度还能再快一点么?我该怎么做呢,将速度压缩在1s秒内呢?只要提升英文的生成速度就好了。十分感谢。
Thank you for the feedback. We have received reports that English inference is slower than Chinese, which is indeed strange. We will investigate the issue and make efforts to improve it. After the modification, how many seconds will be required for English inference?
我找到了解决方法了,我尝试将frontend_en.py中的preprocess_english 模块直接调用,速度提升了许多
感谢您的反馈。我们收到报告说,英语推理比中文慢,这确实很奇怪。我们将调查该问题并努力改进它。修改后,英语推理需要多少秒?
我找到了解决方法了,我尝试将frontend_en.py中的preprocess_english 模块直接调用,速度提升了许多
你好,在修改后,1-3秒,修改前:13-14秒
另外我想咨询一下,如何提升英语模型推理时间
Alright, thank you for providing the information. We will proceed with analyzing the code and will provide you with feedback.
Hit same issue, when using english as input. And used above PLL-L's fix, it work.
The commit with the message "update of frontend, fix bug of slow english speed" (e2da34f61c66896e79b0f411e141189ed22555ad) was made to resolve the issue of slow English processing. Please update and verify if the problem has been resolved.
The commit with the message "update of frontend, fix bug of slow english speed" (e2da34f) was made to resolve the issue of slow English processing. Please update and verify if the problem has been resolved.
Tried it with latest code, issue get fixed. Thanks. Extra: after I git pull latest code with this fix, and report missing pypinyin_dict module in python. and I have to run 'pip install pypinyin_dict' to fix it.
如图,cuda正常![image](https://github.com/netease-youdao/EmotiVoice/assets/69779037/54b7a23a-58cf-4313-86d6-7c1b2c66fcb4)