Gemini 2.5 Pro TTS Language Codes

Gemini API
,
1

Hi

Problem you have encountered:

There seems to be a problem with the Gemini-2.5-pro-tts not recognising language codes.

This is happening in both the production and on the test site versions of the model (link below where this can be reproduced)

https://cloud.google.com/text-to-speech?hl=en

location code set, voice applied and use the default text to generate - it refuses to switch to the correct language code - united kingdom in this case. If you switch models to Gemini-2.5-flash-tts it works.

What you expected to happen:

The voice to speak with a en-gb (united kingdom) voice when generated.

Steps to reproduce:

Using the Gemini-2.5-pro-tts model, use Algenib as the voice, select en-gb as the language code (English (united kingdom)) , let it generate the default text in the demo. Same thing is happening when I try to use the API now.

Other information (workarounds you have tried, documentation consulted, etc):

Tried many work arounds and it does not work - seems to be an issue with the model. This is consistently happening for all voice generations.

1 Like
3

I’m testing voice generator in Playground in AI Studio for a few days now.
In the first day i was able to get perfect Castilian Spanish accent in 2.5 Pro Preview TTS with almost any voice.

The next day I could only get that pronunciation with a few voices.

On the third day I can’t get any voice on any model to sound properly.

Used many variations of “Castilian Spanish accent; language_code: “es-ES” standard, pronunciation” in the style instructions.

Seems broken.

5

Today I tried a different approach.

Instead of a normal text in Spanich I provided the text with some letters written in phonetic alphabet. Specifically θ sound. And TTS could read those “th” sounds properly and consistently with one voice - Autonoe.

image
image1098Ă—709 23.5 KB

It’s super weird :slight_smile: