1. Lightweight deep learning model for assessment of substitution voicing and speech after laryngeal carcinoma surgery