Live Voice - before model callback, automatic-language-switch

[A-b-h-a-y-0-2]

I am trying to apply before_model_callback in an agent, with model as gemini-live-2.5-flash-preview, and i am facing an issue, It works in the text based input for language switch, but for voice output, it returns in the original language[no switch].

What could be the Issue.

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: The before_model_callback modifies the request to the LLM, which is why your text output is in the switched language. However, the voice output is controlled by the SpeechConfig within RunConfig, which is not being updated. To change the voice output language, you need to update the language_code in the SpeechConfig.

---

Hello! I see you're having an issue with changing the language of the voice output. Here's an explanation of what's happening and how you can fix it.

The before_model_callback is designed to intercept and modify the LlmRequest just before it's sent to the language model. [1] By changing the prompt within this callback, you are correctly instructing the LLM to generate a text response in the desired language.

However, the voice output is synthesized by a separate text-to-speech (TTS) engine, and its configuration is managed by the SpeechConfig object within RunConfig. [2] The before_model_callback does not affect this configuration.

To change the language of the synthesized voice, you need to update the language_code property of your SpeechConfig object. This should be done in your application logic before you call the run_live method, by passing a new RunConfig with the updated SpeechConfig.

Here is a conceptual example of how you might adjust the SpeechConfig:

from adk.core import RunConfig, SpeechConfig

# Assuming you have your agent and other configurations set up

# When you need to switch the language to Spanish for voice output
spanish_speech_config = SpeechConfig(language_code="es-ES")
run_config_spanish = RunConfig(speech_config=spanish_speech_config)

# Now, run your agent with the updated RunConfig
agent.run_live(run_config=run_config_spanish)

# To switch back to English voice output
english_speech_config = SpeechConfig(language_code="en-US")
run_config_english = RunConfig(speech_config=english_speech_config)
agent.run_live(run_config=run_config_english)

You can find more details on how to configure streaming and run configurations in the ADK documentation. [3]

I hope this helps you resolve the issue!

---

[1] https://google.github.io/adk-docs/callbacks/types-of-callbacks/
[2] https://google.github.io/adk-docs/runtime/runconfig/
[3] https://google.github.io/adk-docs/streaming/configuration/