Introduction: Computer-aided diagnostic systems with artificial intelligence can enhance healthcare efficiencies. This interactive skill-building session introduces attendees to the Speech Movement and Acoustic Analysis Tracking (SMAAT) software prototype. This application, currently under development, has two primary goals: (1) to derive objective measures of speech articulatory movement, and (2) to generate phonetic transcriptions. These functions are aimed at assisting speech-language pathologists in the assessment of speech sound (SSD).
Method: Audio-visual and perceptual data were collected from 200 children aged 2-4 years and 50 adults who produced the standard set of 40 words and 4 phrases from the criterion-referenced motor speech hierarchy-probe words (MSH-PW). The SMAAT prototype employs the BlazeFace facial-mesh-detector to capture clinically relevant jaw and lip movements, for scoring items on the MSH-PW, such as jaw range and lip rounding/retraction. These movements are time-normalized and standardised to generate norm-referenced spatiotemporal profiles with confidence intervals to support the benchmarking of performance against peers. A transformer-based model automatically extracts the phonetic transcription of each word and phrase for the classification of error patterns. Training examples will be used by attendees to apply the SMAAT app protocol and interpret output.
Results: SMAAT validation and reliability data will be presented, including strong correlations between the expected facial movements and movements extracted for 50 adults with typical speech production, utilising the leave-one-out cross-validation method. Phonetic analysis using a publicly available dataset of dysarthric speech shows our transcription model achieved a 72.70% harmonic mean accuracy. During the workshop, we explore differences in speech-motor control measurements between typical and atypical (SSD) 2-4 year-old children.
Conclusion: Our workshop offers attendees the opportunity to engage in a hands-on demonstration using pre-recorded speech samples to explore the usefulness of spatiotemporal charts and automated phonetic transcription in identifying differences in speech motor control in the assessment of SSD.

At the conclusion of this interactive skill-building session, attendees will take away:
1. Knowledge of how to use the SMAAT app to extract objective measures of speech articulatory movement (i.e., the jaw and lips), within the clinical setting.
2. The capacity to read and interpret objective measures of articulatory movements in children aged 2 years to 4 years of age, and benchmark performance against peers.
3. Use this information, as part of the diagnostic process when assessing speech sound disorders in children aged 2 to 4 years.