Hi, I noticed that gemini series of models can process raw audio. In such cases after user turn completion do we still send transcript in pipeline mode or do we have an option to send raw audio as well? Just curious if that will improve latency or transcript related issues
a
ancient-ability-65292
11/20/2025, 1:17 AM
It does improve lat but you only get one mode
ancient-ability-65292
11/20/2025, 1:17 AM
Either audio or text
ancient-ability-65292
11/20/2025, 1:17 AM
You won’t get the transcribe
m
microscopic-van-26184
11/20/2025, 11:21 AM
I was referring to Gemini 2.5 flash etc. They can take in raw audio but output only text. So I was thinking about if we can skip STT entirely in such scenarios. LLM node directly consumes audio, which outputs text and use a TTS after that. @refined-appointment-81829 thoughts on this?