# Gemini-live-2.5-flash-native-audio Input Transcript generating random languages

**URL:** <https://discuss.ai.google.dev/t/gemini-live-2-5-flash-native-audio-input-transcript-generating-random-languages/138867>\
**Category:** Gemini API\
**Created:** [April 9, 2026, 11:02am UTC](https://discuss.ai.google.dev/t/gemini-live-2-5-flash-native-audio-input-transcript-generating-random-languages/138867 "2026-04-09T11:02:30Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![Anupam\_Mishra](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/anupam_mishra/32/32091_2.png) [@Anupam\_Mishra](https://discuss.ai.google.dev/u/Anupam_Mishra)\
**Post date:** [April 9, 2026, 11:02am UTC](https://discuss.ai.google.dev/t/gemini-live-2-5-flash-native-audio-input-transcript-generating-random-languages/138867/1 "2026-04-09T11:02:30Z")

</div>

I am using the gemini-live-2.5-flash-native-audio with my Google Agent development kit agent. Suddenly, it started generating input transcripts in Hindi, Chinese, and other random languages, even though I had explicitly instructed it in the prompt to use english. I tried disabling VAD and using [ActivityStart](https://ai.google.dev/api/live#activitystart) and [ActivityEnd](https://ai.google.dev/api/live#activityend) instead, but the issue remains the same.

```auto
 run_config = RunConfig(
        streaming_mode=StreamingMode.BIDI,
        response_modalities=["AUDIO"], # Required for your PCM player
        input_audio_transcription=types.AudioTranscriptionConfig(),
        output_audio_transcription=types.AudioTranscriptionConfig(),
        session_resumption=types.SessionResumptionConfig(),
        context_window_compression=types.ContextWindowCompressionConfig(
            trigger_tokens=100000, # Start compression at ~78% of 128k context
            sliding_window=types.SlidingWindow(
                target_tokens=80000 # Compress to ~62% of context, preserving recent turns
            )
        ),
        proactivity=types.ProactivityConfig(proactive_audio=True) if proactivity else None,
        enable_affective_dialog=affective_dialog,
        speech_config=types.SpeechConfig(
            voice_config=types.VoiceConfig(
                prebuilt_voice_config=types.PrebuiltVoiceConfig(
                    voice_name=os.getenv("AGENT_VOICE", "Puck")
                )
            ),
            language_code=os.getenv("AGENT_LANGUAGE", "en-US")
        )

```

Any idea, how can this be fixed.

---

<div class="post-metadata">

**Author:** ![Mustan\_lokhand](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/mustan_lokhand/32/86726_2.png) [@Mustan\_lokhand](https://discuss.ai.google.dev/u/Mustan_lokhand)\
**Post date:** [April 9, 2026, 9:57pm UTC](https://discuss.ai.google.dev/t/gemini-live-2-5-flash-native-audio-input-transcript-generating-random-languages/138867/2 "2026-04-09T21:57:01Z")

</div>

Hi @Anupam_Mishra , Does this happen on every run or is this a sporadic issue ?

---

<div class="post-metadata">

**Author:** ![Anupam\_Mishra](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/anupam_mishra/32/32091_2.png) [@Anupam\_Mishra](https://discuss.ai.google.dev/u/Anupam_Mishra)\
**Post date:** [April 10, 2026, 4:06am UTC](https://discuss.ai.google.dev/t/gemini-live-2-5-flash-native-audio-input-transcript-generating-random-languages/138867/3 "2026-04-10T04:06:00Z")

</div>

Almost every time. It will suddenly pick one or two word’s and respond in a random language.

I have seen this is also notified by other users as well for the preview model. But for me the issue appears in Gemini-live-2.5-flash-native-audio also.

Although what I said is correctly identified by the api. Only the input transcript generated is in some random language.

[https://discuss.ai.google.dev/t/gemini-live-api-gemini-2-5-flash-native-audio-preview-12-2025/119862/5](https://discuss.ai.google.dev/t/gemini-live-api-gemini-2-5-flash-native-audio-preview-12-2025/119862/5)

[https://discuss.ai.google.dev/t/input-transcription-in-gemini-live-api-is-very-weird/112644](https://discuss.ai.google.dev/t/input-transcription-in-gemini-live-api-is-very-weird/112644)