Speech transcript endpoint returning empty text array for completed calls

Calling the endpoint to retrieve the transcript for a specific conversation returns a 200 OK, but the text array in the JSON payload is always empty. The transcript status shows COMPLETED and duration is populated correctly. Tried waiting 5 minutes between calls. Here is the response :

{
 "transcriptionId": "a1b2c3d4-...",
 "status": "COMPLETED",
 "text": []
}

Is there a specific flag needed in the initial transcription request to populate the text data?