Text to speech voices sounding different?

[Meeps]

Am I going insane? Did they change the Wavenet voices for Text to Speech? This past week I have noticed the voices have been lower quality and they sound vastly different. I've talked to a few friends who confirmed my findings. What's going on? Am I doing something wrong? Was there a change? Why didn't they announce a change? Or did they announce it and I missed it. Are the servers somehow overloaded and they are serving lower quality audio? I'm just confused. Users generally a very tied to how a voice sounds and the recent change has been upsetting for many of those people.

[lsolatorio]

Hi @Meeps,

Thank you for joining our community.

I completely understand how frustrating it must be to have your Text-to-Speech voice behave unexpectedly. It sounds like you've put some effort into troubleshooting already, and I apologize that there aren't any official announcements or status health reports available.

Here's what I found that might be related:

1. Another post in our community titled "Text-2-Speech voice has been changed", unfortunately, there aren't any confirmed solutions yet.
2. There's also an open issue submitted on February 14, 2024 named "ja-JP-Wavenet-A Request Outputs ja-JP-Wavenet-B Voice" (you may not have access to the details of the ticket)

You can check the currently supported wavenet voices to confirm if there are changes but it's best to reach out to Google Cloud Support for better assistance or submit another issue ticket provided with the specifics of your use case.

I hope I was able to provide you with useful insights.

[smarty]

Hello,

I have been experiencing what appears to be the same problem since yesterday, about 20 hours ago.

I have been using Wavenet voices through the API for years without incident.

The Wavenet and Neural2 voices normally sound much more life like but now they sound extremely monotonous & much more robotic which is unacceptable for my users.

It's easy to reproduce the problem using the Demo:

https://cloud.google.com/text-to-speech

If you select English (Great Britain) Wavenet and any voice name and press the Speak It button then you will hear that it sounds flat, like the Basic voice type. 

Meanwhile other voice types such as Journey are working normally.

I have checked that I was not experiencing this problem when @Meeps  posted. @Meeps did this problem go away?

Has anyone else able to reproduce the behaviour?

[paul-biped]

Same issue here, observed with Wavenet (specifically fr-FR-Wavenet-E)

As I still have local files that were generated through gcloud TTS, I could pinpoint the change as happening between 17.01.2025 and 29.01.2025

The assets that are newly generated do match the demo link provided by @smarty for me, and don't match the examples from the documentation https://cloud.google.com/text-to-speech/docs/voices
Is that a bug, or is the change of quality on the wavenet voices to stay ?

[paul-biped]

Actually found a relevant issue ticket here https://issuetracker.google.com/issues/392651795