Repository navigation
Audio transcription not returned in response when using gemini-live-2.5-flash-preview-native-audio and following documentation #1279
Description
Hi,
I'm trying to use the Vertex AI Gemini Live API to get audio transcription for both input and output audio, following the official documentation:
https://cloud.google.com/vertex-ai/generative-ai/docs/live-api?hl=pt-br#transcribe_audio
I'm using the recommended model for audio transcription:
gemini-live-2.5-flash-preview-native-audio
My configuration is as follows:
# config.py
if api_config.use_vertex:
MODEL = os.getenv('MODEL_VERTEX_API', 'gemini-live-2.5-flash-preview-native-audio')
else:
MODEL = os.getenv('MODEL_DEV_API', 'models/gemini-live-2.5-flash-preview-native-audio')
CONFIG = {
"generation_config": {
"response_modalities": ["AUDIO"],
"output_audio_transcription": {},
"input_audio_transcription": {},
"speech_config": VOICE,
},
"system_instruction": SYSTEM_INSTRUCTIONS
}
I send audio data to the model like this:
# websocket_handler.py
async for message in websocket:
data = json.loads(message)
if data["type"] == "audio":
await session.genai_session.send(input={
"data": data.get("data"),
"mime_type": "audio/pcm"
}, end_of_turn=True)
And I process the response as follows:
# websocket_handler.py
if data["type"] == "audio":
await session.genai_session.send(input={
"data": data.get("data"),
"mime_type": "audio/pcm"
}, end_of_turn=True)x
Problem:
Even though I am using the correct model and configuration, and sending audio as described in the documentation, the input_transcription and output_transcription fields in the response are always missing.
I have tried with both Vertex and Dev endpoints, but the transcription is never returned.
Environment:
Model: gemini-live-2.5-flash-preview-native-audio
Endpoint: Vertex AI (VERTEX_API=true)
Python SDK: google-genai
Audio format: PCM, 16kHz
References:
Vertex AI Gemini Live API - Transcribe audio
Vertex AI Gemini Live API - Output fields
Expected behavior:
When sending audio, the response should contain the transcription in input_transcription.text and/or output_transcription.text as described in the documentation.
Actual behavior:
The transcription fields are always missing.
Questions:
Is there any additional configuration required to enable audio transcription?
Is this a known issue with the current model or SDK?
Is there any way to debug or inspect why the transcription is not being returned?
Thank you for your help!
Activity
Braga-Pedro commentedon Aug 19, 2025
I am trying with the models:
But not success.
shivvaam0001 commentedon Sep 24, 2025
Hi @Braga-Pedro
We heard your feedback and We have released the new
gemini-2.5-flash-native-audio-preview-09-2025model. Could you please try this and check if it resolves your issue with the new model and let us know what you think?Thank you for your patience.
Braga-Pedro commentedon Oct 9, 2025
thanks for reply @shivvaam0001 , i tested with the gemini-2.5-flash-native-audio-preview-09-2025 model but ir returned this erro:
2025-10-09 11:44:56,064 - ERROR - Error establishing realtime session: received 1008 (policy violation) Permission 'aiplatform.endpoints.predict' denied on resource '//aiplatform.googleapis.com/projects/andy-voice-470817/locati; then sent 1008 (policy violation) Permission 'aiplatform.endpoints.predict' denied on resource '//aiplatform.googleapis.com/projects/andy-voice-470817/locati
jval7-aceup commentedon Oct 14, 2025
I'm experiencing the same problem. Do you have any updates?
Braga-Pedro commentedon Oct 15, 2025
I havent had an updates yet ;(
jval7-aceup commentedon Oct 15, 2025
@Braga-Pedro Hi try with >=1.43.0 version, it worked for me with gemini-2.5-flash-native-audio-preview-09-2025
I have identified previus version does not have the field for the transcripts
abhinav-TB commentedon Oct 28, 2025
same issue here
shivvaam0001 commentedon Nov 13, 2025
Hi @Braga-Pedro, @abhinav-TB
Could you please try using version 1.43.0 and let me know if you are still experiencing the same issue?
abhinav-TB commentedon Nov 14, 2025
@shivvaam0001 , it worked for me after a couple of days
github-actions commentedon Nov 22, 2025
This issue has been marked as stale because it has been open for 7 days with no activity. It will be closed in 2 days if no further activity occurs.
github-actions commentedon Nov 24, 2025
This issue was closed because it has been inactive for 9 days. Please post a new issue if you need further assistance. Thanks!
sbenodiz commentedon Jan 3, 2026
I have the same issue, any updates ?
Marincor commentedon Jan 7, 2026
any updates?
shivvaam0001 commentedon Jan 13, 2026
Hi @sbenodiz & @Marincor
Since this issue has been resolved, could you share some additional details about your issue?
Thanks!
sbenodiz commentedon Jan 13, 2026
It works for me! Thanks
fix(orb): Use correct Gemini model with transcription support
VTID-01225: Fix Cognee memory context bootstrap (#501)