SPEECH · TECHNOLOGY · EQUITY

Objective SSD
identification for
every child.

A machine learning model for pediatric speech sound disorder identification, built by a clinician, grounded in 54 SLP interviews, and designed for the real world.

Request a conversation Learn more

SLP interviews conducted across every practice setting: school, clinic, hospital, and early intervention

7M+

annual U.S. SSD evaluations, representing a $259 to $504M market with no objective acoustic tool

Zero

objective, acoustic-based SSD identification tools currently in clinical or school use

01Speech sound disorders affect ~8% of U.S. children

02107% increase in SSD diagnoses since 2020

0327% of school SLPs considering leaving the field · ASHA 2024

041 to 2 hours per evaluation, scored by hand

05GFTA-3 normed on just 25 samples per age cell

06LLC Est. 2023 · Harvard-affiliated MS in SLP

01Speech sound disorders affect ~8% of U.S. children

02107% increase in SSD diagnoses since 2020

0327% of school SLPs considering leaving the field · ASHA 2024

041 to 2 hours per evaluation, scored by hand

05GFTA-3 normed on just 25 samples per age cell

06LLC Est. 2023 · Harvard-affiliated MS in SLP

Mission

Why TALK exists.

Speech sound disorders affect approximately 8% of children, yet diagnosis remains subjective, variable, and inequitably distributed. The tools are outdated. The burden falls on clinicians. The children bear the gap.

01 ——

Reduce evaluator variability

SSD diagnosis varies across evaluators, settings, and experience levels. Acoustic-grounded analysis makes the outcome consistent, regardless of who performs the evaluation.

02 ——

Expand access to accurate diagnosis

Children in under-resourced schools and underserved communities face the longest waitlists and the most inconsistent care. Objective tools change that calculus at scale.

03 ——

Advance the evidence base

Research findings will be shared openly with the field. TALK is committed to contributing to, not just consuming from, the science of pediatric speech-language pathology.

The numbers

Need documented.
Gap confirmed.

Prevalence

~8%

of U.S. children have a speech sound disorder, with diagnoses up 107% since 2020.

Annual evaluations

7M+

SSD assessments conducted annually in the U.S., at $37 to $72 per evaluation, totaling a $259 to $504M annual market.

Primary research

in-depth SLP interviews conducted across all practice settings, validating the problem and informing every product decision.

If this tool could squeeze time from 1.5 hours into 20 to 30 minutes, that would be helpful. I could spend that time somewhere else. School-based SLP, Jordan School District (primary research interview)

The problem

Current tools are
incomplete by design.

27%

of school-based SLPs are considering leaving the field. Burnout is the leading cause. Paperwork and manual scoring are the leading contributors. Source: ASHA 2024 Annual School Survey.

1 to 2 hrs

per child for evaluation and scoring, not including report writing, which can add another 45 to 90 minutes. Most SLPs report working unpaid overtime weekly to keep pace.

samples per age cell, the norm base for the GFTA-3, the current gold-standard assessment tool. Administered with pen and paper. Nearly every SLP interviewed described it as incomplete.

Clinical listening alone was never enough.

Speech sound disorder identification is inherently subjective. Two trained clinicians can evaluate the same child and reach different conclusions. The GFTA-3, the dominant tool for decades, was normed on a small, homogeneous sample and requires manual elicitation, transcription, and analysis by hand.

Across 54 SLP interviews, the consistent finding: too much time on scoring, no objective verification, no automated pattern detection, and no clear path from evaluation data to diagnosis that does not run entirely through the clinician's judgment and memory.

TALK is building the missing infrastructure: an acoustic ML model that classifies speech sound errors objectively, acts as a second set of ears, and integrates into the workflows SLPs already use.

The research

Building the model.

A supervised ML approach to acoustic SSD identification

TALK is developing a supervised machine learning model to identify acoustic markers associated with speech sound disorders in children. The model will be trained and validated on clinically annotated pediatric speech data, with the goal of producing a tool that integrates into existing SLP evaluation workflows.

Access to high-quality, annotated pediatric speech corpora is critical to this work. TALK is actively seeking research data partnerships with institutions whose existing datasets can help ground the model in clinically valid, diverse speech samples.

Acoustic feature extraction from pediatric speech samples
Supervised classification of speech sound error patterns
Cross-evaluator validation and inter-rater reliability analysis
Differentiation between disorder, difference, and typical variation
Built by a CCC-SLP with direct clinical SSD evaluation experience

About the founder

Clinician. Technologist. Builder.

Heidi Blackham

MS • CCC-SLP • Engineering Program Manager • Founder & CEO

I am a licensed, ASHA-certified speech-language pathologist with a master's degree from a Harvard-affiliated SLP program, and I have spent the last several years building the case that clinical rigor and technical ambition are not just compatible, they are the only combination that actually solves this problem.

The idea for TALK came while I was practicing in Boston, sitting across from children and their families in evaluation rooms, running pen-and-paper assessments that had not meaningfully changed in decades. I knew the tools were inadequate. I also knew that the data, the computing infrastructure, and the machine learning frameworks to do something about it all existed. They had simply never been brought together with genuine clinical depth.

Since forming TALK LLC in 2023, I have conducted 54 in-depth interviews with SLPs across every practice setting to validate the problem and inform the product. I have assembled an advisory team across machine learning, software development, business strategy, and clinical practice. I also serve as an engineering program manager at a national technology company, giving me direct experience leading complex software programs from concept to production.

What I am building is not a clinical decision-support chatbot. It is a validated acoustic ML model, built with the same rigor that SLPs apply to their evaluations, because anything less would not be worth building.

CCC-SLP Harvard-affiliated MS Pediatric SSD evaluation Engineering PM 54 SLP interviews ASHA member Advisory team assembled

2021

Idea born while practicing as an SLP in Boston

2023

TALK LLC formed; transitioned into tech

2025

54 interviews complete; advisory team assembled; data partnerships underway

FAQ

Honest answers.

TALK is developing a supervised machine learning model that uses acoustic analysis to identify speech sound disorders in children. The goal is an objective, data-grounded reference tool that supports speech-language pathologists during evaluation, reducing variability, saving time, and expanding access to consistent diagnosis.

No. The model is designed as a clinical decision-support tool, not a replacement for the clinician. SSD identification is complex and context-dependent, and it involves nuance that only a trained SLP can fully evaluate. TALK's role is to make the acoustic analysis side of that evaluation objective and faster, so clinicians can spend more time where it matters most.

The long-term vision is a validated, clinically integrated tool deployed across school-based and clinical settings, with published findings, open contributions to the research community, and expanded capabilities including differentiation between disorder and linguistic difference, multilingual populations, and sentence-level analysis.

TALK was founded by Heidi Blackham, MS, CCC-SLP, a licensed speech-language pathologist with a master's degree from a Harvard-affiliated SLP program and a current role as an engineering program manager at a national technology company. TALK LLC was formally established in 2023. An advisory team with expertise across machine learning, software development, business strategy, and clinical practice supports the work.

Objective SSD identification for every child.