How to integrate SpeechClient V2 for bidirectional streaming voice applications in LLMFlow

[ammmr]

Thank you for creating this excellent Agent Development Kit.

I would like to create a real-time bidirectional streaming voice application using SpeechClient V2, but I noticed that adk-python is using V1 in the audio_transcriber.py file:
https://github.com/google/adk-python/blob/main/src/google/adk/flows/llm_flows/audio_transcriber.py

Could you please advise on how to easily integrate SpeechClient V2 into LLMFlow for bidirectional streaming support?

Thank you in advance for your help and for maintaining this great repository.

Best regards,

[hangfei]

We will switch to a built-in transcription so this transcriber will not be needed.

Any features you are expecting from SpeechClient V2?

[ammmr]

@hangfei

Thank you for your response. If you're planning to switch to a built-in transcription solution, I'm happy to wait for that implementation rather than modifying the current transcriber.

My main interest in SpeechClient V2 was to leverage the Chirp2 model for Speech-to-Text functionality. I was hoping to create a real-time bidirectional streaming voice application that could benefit from Chirp2's improved accuracy and performance.

Would the upcoming built-in transcription solution potentially support Chirp2 or similar advanced speech recognition models? And do you have any timeline for when this built-in solution might be available?

Thanks again for your work on this project.

[chulkilee]

What would be "built-in transcriber"?

I'm wondering whether it will allow integrating non-google STT/TTS services.