Fig. 1: The overall framework for diagnosing autism and identifying language disorders.

The conversation is first passed to stages of speaker diarization and audio transcription. Then we designed two prompts for GPT-based Models to generate a response parser. The generated response parser was mapped to an A4 Score (based on domain knowledge from ADOS-2 Module 4): Stereotyped/Idiosyncratic Use of Words or Phrases (see Table 3).