The timbre of the voice of "zh-CN-XiaochenNeural" has changed,why does it seem like a different person?

Question

The timbre of the voice of "zh-CN-XiaochenNeural" has changed, and the current timbre is completely different from what it was a few weeks ago. The results of the test using "Batch synthesis API (Preview) for text to speech" or "Speech Studio" were different from those of a few weeks ago.My SSML input would look something like this:“

这天的最后，是辅导员带我去校医务室做了伤口处理。她叹着气坐在病床旁边，看我的眼神充满了怜悯。你们的事我都知道了，老师相信，这次也不是你的错......”。

I still have a voice mp3 file generated a few weeks ago locally, and I can't generate the same voice as a few weeks ago. Are there any adjustments made to the "zh-CN-XiaochenNeural" voice pack?

Batch synthesis API (Preview) :

https://eastasia.customvoice.api.speech.microsoft.com/api/texttospeech/3.1-preview1/batchsynthesis/

2023/11/22

The new finding is that the results obtained by using the Microsoft Speech API are correct, but this is a free presentation service that is unstable.

Microsoft Speech API:

https://southeastasia.api.speech.microsoft.com/accfreetrial/texttospeech/acc/v3.0-beta1/vcg/speak

So why does zh-CN-XiaochenNeural in Azure Speech API(Batch synthesis API (Preview)) suddenly become a different voice compared to Microsoft Speech API?

Answer

This is solved by technical support. Here are the results:

问题描述：
您正在使用Azure文本转语音服务。您的语音资源部署在东亚区域（East Asia），在使用“zh-CN-XiaochenNeural”模型时遇到了音色（timbre）与之前表现不同的情况。并且，您分别提供了该模型几周前正确的输出音色样本（good case）与目前错误的输出音色样本（bad case）供参考。

调查反馈：
经过测试，我们确定了东亚区域语音资源的“zh-CN-XiaochenNeural”模型音色存在改变的情况，并正在积极地与产品团队商讨针对这一区域语音资源的解决措施。

缓解建议：
您可以将您的语音资源切换至东南亚区域（Southeast Asia），该区域“zh-CN-XiaochenNeural”模型的音色与您提供的good case音色一致。

Share via

The timbre of the voice of "zh-CN-XiaochenNeural" has changed,why does it seem like a different person?

1 answer