ChiReSSD reconstructs the speech of children with speech sound disorders, suppressing mispronunciations while keeping the child's voice and prosody: 48% relative CER reduction and …
MMDyS: severity-conditioned contrastive alignment treats speech–text mismatch as a continuous signal of impairment, fused with hemiface asymmetry through a mixture of experts.
Text-guided diffusion that generates pediatric facial anomalies from text prompts for clinical research.
Adapts TTS and voice conversion to generate synthetic disordered child speech for training pediatric ASR.
A transformer-based lip normalization filters out abnormalities while preserving individual anatomy; a Siamese CNN compares pre- and post-normalization images, trained without …