Gemini 2.5 Flash/Pro Preview TTS – Wrong voice is used + style instructions are ignored

Google AI Studio
1

In Google AI Studio (generate-speech tab) using gemini-2.5-flash-preview-tts or gemini-2.5-pro-preview-tts:

  • I select a specific voice (male/female,) from the dropdown

  • The generated audio almost always comes out in a completely different voice (very often a US-accent female voice it suddenly comes out in a male voice or vice versa)

  • Style instructions added to the prompt (“speak in a happy tone”, “use a sarcastic voice”, “read this like a children’s storyteller”, “sound excited and energetic”, etc.) have no effect – the output stays flat and neutral every time

This started right after the December 10 update that replaced the previous preview TTS models. The issues are 100% reproducible on every single generation.

Thank you!

2

today only one line of words work the rest are ignored

3

Well, something strange is happening i put one text and the text to speech rewrites it ??? comes out complete different narration than the text provides??

4

Yep, they broke the whole model. I get nothing, only silence. Before the update I was able to get 5, 10 and 15 minutes of audio chunks. Now there is nothing left of this quality.

3 Likes
6

Hi @Yoyo2 ,

Could you please provide full prompt along with the generated audio sample that demonstrates this issue? This will help us reproduce the issue on our end.

2 Likes