Arabic Transcription Issue with OpenAI Realtime API

I’m using the OpenAI realtime API with the following session update configuration to enable user input transcription:

"input_audio_transcription": {
  "model": "whisper-1"
}

However, when I speak in Arabic, the final transcript (from events like conversation.item.input_audio_transcription.completed) is sometimes returned in English.

Has anyone encountered this issue with the realtime API or found a fix? Thanks!

1 Like

Hi @eng.hello

You can fix it simply by adding a language param to the input_audio_transcription object.

2 Likes