I’ve been experiencing noticeable latency issues with the Gemini Live API over the past week. When I speak, the input audio processing and the model’s response have become significantly slower, resulting in a poor real-time experience. This issue occurs across both the latest and some older models — previously, everything was working smoothly. The model I have been using is: gemini-2.5-flash-native-audio-preview-09-2025
Also, I tried with the half cascade model but facing the same issue: gemini-2.0-flash-live-001 and gemini-live-2.5-flash-preview
Additionally, when I speak in English, the transcribed input sometimes appears in Hindi. This issue has been observed for quite some time now with the models above.
It would be great if the team could look into:
The recent increase in latency for input during live sessions
The incorrect language detection/transcription (English audio → Hindi text)
Can you share your tier and some example prompts which caused English to Hindi translation issue. It helps us to reproduce and share the feedback to concerned team.
I am using the Free tier api key as of now for the Live API.
Regarding the language issue, I have attached screenshots showing the input transcript text that appears in Hindi, even though my questions are spoken in English.
I’m experiencing the same issue. I’m using the default AI Studio code without changes, and after 1–2 minutes of conversation the latency increases sharply. It starts out normal, but the delay worsens as the session continues. I’ve attached the code and a video for reference.
I experienced high latency as well. I used aistudio to test the voice experience. The latency got worse as my speech got longer. I sometimes waited over 10 seconds before I started to hear something from the model. Here are some of my speech that caused such latency. I got these recordings from aistudio. My tier is 3. Somehow the recording has long pause before and after my actual speech.
Hi @Krish_Varnakavi1 we are at Tier-3 and we are still facing the latency issue as of today 14 november 18:57pm with gemini-2.5-flash-native-audio-preview-09-2025.
Initially the model latency is fine but as the conversation progress after 1 minute the latency increases a lot. Please look into this matter as we are using this model intensively for our business.
