-
作者您好,根据您的代码,我用标贝女生训练了180000步,目前合成的语音质量已经很高了,但是多音字问题发音比较严重,比如"长大"读成chang大,等问题,另外就是不能合成阿拉伯数字。因为刚刚接触语音合成,很多事情不懂,如果特来请教,如果我想优化解决上面两个问题,请问我需要从哪些角度通过什么方式去解决呢?感谢~~!!!
-
默认的 PaddleSpeech fs2 的前端是不含韵律信息的 ,如果我引入了韵律标注, 如何在训练过程中引入这个信息呢? 这是否意味着要修改声学模型的代码? 声学模型的输入pitch energy duration的那些特征信息吧,我该怎么引入韵律信息呢?
SG-XM updated
11 months ago
-
我在用AiShell3的数据集进行训练的时候发现有不少中文在Bert词表中没有,在实际的生产部署时候碰到这种情况应该如何处理?
例如:SSB03930415.wav 搜狐娱乐讯**渀**港媒报道
-
您好,想请教下标点对模型的影响。分别试了一下带标点与不带标点的文本作为text_prompt。发现有标点的生成语音停顿会好些。想问下,是因为在训练时音频对应的文本也是有标点的吗?
-
大大您好,我似乎没看到您在工程里有关于如何用AISHELL3去实现工程的PIPLINE,可能其他大佬们一看都知道如何去部署,但是我刚接触这些问题,是初学者,不知道如何去训练,同时又听说您的团队做的wetts非常的先进,很想尝试学习一下,您方便提供一下关于如何用AISHELL3去部署wetts的Reademe吗?感谢感谢!
-
https://vallex-demo.github.io/
Title
-
环境:
mac m1
`Building for simulator (x86_64)
SHERPA_ONNXRUNTIME_LIB_DIR: /Users/yaoyajie/open-source/sherpa-onnx/build-ios/ios-onnxruntime/onnxruntime.xcframework/ios-arm64_x86_64-simulator
`
e…
-
大佬有打算后续做下ZeroShot的工作嘛?或者有了解过ZeroShot目前性能怎样啊?
-
**Summary[问题简述(一句话)]**
(Tensorflow) E:\ai\demo_toolbox\MockingBird-main>python E:\ai\demo_toolbox\MockingBird-main\demo_toolbox.py
Arguments:
datasets_root: None
vc_mode: …
-
cuda 信息:
```
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2020 NVIDIA Corporation
Built on Tue_Sep_15_19:10:02_PDT_2020
Cuda compilation tools, release 11.1, V11.1.74
Build cuda_11…