Non-English text shown on transcription console UI

[asadullahnaeem-techtics]

The logs of the Multimodal agent with OpenAI Realtime API show correct English text on the console but on the UI, it sometimes shows the audio transcription in other languages like Hindi, Chinese, Russian etc while speaking English.

[longcw]

I noticed the same issue sometimes. According to the openai's document, the transcription is from a separate process run on a separate ASR model (currently whisper-1), and the transcript may diverge somewhat from the model's interpretation. So it seems it's from the speech recognition error. https://platform.openai.com/docs/api-reference/realtime-server-events/conversation/item/input_audio_transcription/completed

[longcw]

oh wait, you mean in the transcripts in logs it's correct but in UI it's wrong? maybe it's a different issue from what I mentioned above.

[asadullahnaeem-techtics]

In the logs, transcription is correct but on the UI it is not.

[davidzhao]

can you share an example with logs or screenshot?