Persian model (vosk-model-small-fa-0.5) transcribes only 1–2 words from full speech

Hi,
I'm testing the Persian model vosk-model-small-fa-0.5 for an offline voice-to-text task. I noticed that the model performs very poorly on clear Persian audio input. Even when the input is a complete sentence spoken clearly in a quiet environment, the output typically includes only one or two unrelated words, or sometimes nothing at all.