Developing an AI-powered speech annotation and transcription enhancer tool

The National Institute on Deafness and Other Communication Disorders awarded Auspex Medix, LLC $305,923 on July 1, 2026, under the Research Related to Deafness and Communication Disorders program (CFDA 93.173) to develop SATE (Speech Annotation and Transcription Enhancer), an AI-powered platform that automates language sample analysis for speech-language pathologists.

SATE automates multi-layer linguistic annotation across word- and phoneme-level transcription, C-unit segmentation, phonetic analysis including mispronunciation detection, maze identification, morpheme and syllable analysis, and grammar evaluation. The system delivers interactive reports with dynamic visualization and real-time editing capabilities, addressing the current workflow bottleneck in which manual annotation requires 7–8 minutes per minute of speech. The platform leverages a proprietary dataset and multiple public corpora (ENNI, L2-ARCTIC, RESCORLA, ULTRASUITE) to ensure robustness across linguistic accents, dialects, and demographic groups.

Phase I will validate technical performance on typical and disordered speech, targeting 0.8 sensitivity and specificity at 95% confidence interval, and assess usability and adoption through participatory studies with 20 speech-language pathologists and heuristic evaluation by 8 human-computer interaction experts. Success criteria include robust annotation accuracy, strong usability ratings, and preliminary evidence of financial feasibility.

Performance occurs in Amherst, New York. The award period runs through June 30, 2027. This is a Project Grant, the assistance type most common in NIH research funding.

Abstract

Language sample analysis (LSA) is the gold standard for speech-language pathologists (SLPS) to evaluate functional communication, but manual annotation is slow and inconsistent, taking 7-8 minutes per minute of speech. This bottleneck delays interventions and exacerbates workforce strain, with 55% of SLPS reporting excessive caseloads. Existing tools such as salt, clan, and whisper offer only partial automation and lack comprehensive, multi-layer analysis. With the U.S. Speech therapy market valued at $4.62 billion in 2023 and projected to grow to $8.37 billion by 2032 (cagr 6.9%), there is strong demand for efficient, scalable, and clinically useful technologies that address SLP workflow needs. We propose sate (speech annotation and transcription enhancer), an AI-powered platform that automates multi-layer LSA with high accuracy and usability. Sate delivers precise annotations across word- and phoneme-level transcription, C-unit segmentation, phonetic analysis (e.g., mispronunciation detection), maze identification (pauses, repetitions, filler words), morpheme and syllable analysis, and grammar evaluation. Outputs are presented in interactive reports with dynamic visualization and enable real-time editing, enabling efficient review and refinement for SLPS. The system leverages both a proprietary dataset and multiple public corpora (e.g., enni, L2-arctic, rescorla, ultrasuite) to ensure robustness across linguistic accents, dialects, and demographic groups. Phase I will validate technical performance on typical and disordered speech, targeting ≥0.8 sensitivity and specificity (95% CI), and assess usability and adoption through participatory studies with 20 SLPS and heuristic evaluation by 8 HCI experts. Success will be defined by robust annotation accuracy, strong usability ratings, and preliminary evidence of financial feasibility. By reducing the time, effort, and variability of LSA, sate will expand service capacity, alleviate clinician burden, and improve access to high-quality speech and language evaluation. Completion of phase I will provide the technical, clinical, and commercial foundation for phase II validation, workflow integration, and saas-based commercialization, positioning sate to transform LSA and strengthen speech and language service delivery.