AI speech therapy for dysarthria.

A'ber is an AI powered clinical speech therapy platform designed specifically for dysarthria. It covers all dysarthria etiologies including hypokinetic dysarthria in Parkinson's disease, ataxic, spastic, flaccid, and mixed types. The platform uses ML based severity classification to assess speech and a five layer adaptive engine to personalize therapy across 8 speech sub tracks: articulation, phonation, respiration, resonance, intonation, stress, speech rate, and loudness.

How A'ber works

A'ber begins with a short acoustic assessment. The patient completes a battery of nine speech tasks, including sustained vowel phonation, diadochokinesis, sentence projection, pitch contour glides, and a reading passage. From these recordings a machine learning classifier estimates severity at the dimension level rather than producing a single global score. The clinical engine then maps the severity profile onto a tailored therapy plan.

Sessions follow a six step rhythm. The patient begins with a brief warm up, then progresses through three to five targeted exercises selected by Thompson Sampling against the active sub track. After each exercise the system computes acoustic metrics (jitter, shimmer, cepstral peak prominence smoothed, normalized pairwise variability index, articulation rate, maximum phonation time, intensity) and updates a Kalman filtered estimate of the patient's skill on that dimension. The session closes with a pre and post probe and a one screen recap of what changed.

Between sessions the adaptive engine uses Elo difficulty calibration to adjust task difficulty for the next session, and Bayesian cross patient pooling to warm start patients whose data is still sparse. Nothing about the protocol is generic. Every selection is grounded in the patient's own acoustic history and the population prior for similar severity profiles.

Who A'ber is for

A'ber is built for adults living with dysarthria from any etiology, including Parkinson's disease, stroke, traumatic brain injury, ALS, multiple sclerosis, and cerebral palsy. It is also built for the speech language pathologists (SLPs) who treat them. On the patient side, A'ber turns daily practice into a ten minute routine that runs at home with clinical grade feedback after every recording. On the clinician side, A'ber extends one SLP's reach from a typical eight patient caseload to thirty or more without sacrificing per patient quality, because the adaptive engine handles exercise selection and intensity decisions that previously consumed an SLP's clinical time.

The clinical foundation

The exercise protocols are derived from the motor learning literature on dysarthria rehabilitation. A'ber follows the knowledge of performance to knowledge of results transition recommended by the Schmidt and Lee framework, providing rich feedback early in a new skill and faded summary feedback once the patient is consolidating. Feedback frequency is calibrated per sub track because the literature shows that the optimal cadence differs for respiration relative to articulation. Every exercise carries an explicit clinical rationale that links it back to the motor speech subsystem it targets.

How A'ber compares to general speech therapy apps

Unlike general purpose speech therapy apps that cover aphasia, cognitive rehabilitation, voice, and pediatric articulation in one product, A'ber focuses exclusively on the motor speech deficits that define dysarthria. The narrower clinical scope is what allows the platform to ship a severity classifier and an adaptive engine that are tuned to dysarthric acoustics, not adapted from a general language therapy template. For SLPs already running LSVT LOUD or a similar protocol, A'ber is a complement, not a replacement: it extends practice between clinician visits and keeps the daily intensity that the evidence calls for.