Apps That Score a Child's Pronunciation, and What the AI Can and Can't Hear

A few years ago, the only way to know whether your child said "rabbit" or "wabbit" correctly was to listen yourself and hope your ear was right. Now a handful of apps will listen for you, put a number on the attempt, and sometimes tell you which sound slipped. That is genuinely useful. It is also easy to over-trust, because a score of 84 looks a lot more certain than it is.

Here's the TLDR: automatic pronunciation scoring is good at one job: giving a child more practice attempts with some feedback when an adult isn't sure what they heard. It is not good at telling you whether a speech problem is fixed, and no app we looked at can reliably catch distortions (a slushy lateral lisp, an R that's "almost there"). Treat every score as practice feedback, not an assessment, and keep your child's speech-language pathologist (SLP) as the judge.

If you just want the shortlist:

1. Speech Therapy Magic: best for sound-by-sound scoring of speech-therapy target words in a browser. $9.99/mo or $79.99/yr after a 7-day trial (no card).
2. Google Read Along: best free option for a young child who is learning to read aloud. It listens to reading, not to articulation targets.
3. Microsoft Reading Coach: best free option for older readers who stumble on specific words.
4. ELSA Speak: best for teens working on an English accent. It is built for language learners, not for speech therapy.
5. Articulation Station Hive: no automatic scoring at all, and that's the point. A grown-up taps right or wrong, which is the most accurate "scoring" you can get at home.

Prices checked October 2026. The rest of this guide explains how these tools listen, what the research says about AI and children's speech, and how to use a score without letting it mislead you.

How Pronunciation Scoring Actually Works

Ordinary speech recognition (the kind in Siri or a dictation app) is built to figure out which words you said, and it's forgiving on purpose: if a child says "wabbit," it will happily type "rabbit." That's the wrong behavior for speech practice.

Pronunciation scoring flips the question. The app already knows the target word, so it lines the audio up against the sounds that word should contain (r-a-b-i-t) and asks, sound by sound, how closely each stretch of audio matches the expected sound. Researchers call this goodness of pronunciation. The output is usually a score per sound, rolled up into a word score, and sometimes a guess at what the child said instead ("heard /w/").

That design has a built-in limit worth understanding before you trust any app. The model can only choose between the sounds it knows. A clean swap of one sound for another (/w/ for /r/, /th/ for /s/) is something it can at least try to name. A distortion, a sound that isn't cleanly any English sound, has no label to land on, so the model often scores it as close enough. That is why lateral lisps and partly-correct R sounds are the hardest cases for every tool here.

What the Research Says About AI and Children's Speech

This isn't a new problem, and researchers have been honest about it for years:

• A systematic review of automated speech analysis tools for children (McKechnie et al., 2018, International Journal of Speech-Language Pathology) found that tools often reached 80% agreement with human judges for broad questions like overall intelligibility or severity, but that accuracy was "typically <80%" for words that actually contained mispronunciations, which is exactly where a speech-practice app needs to be right.
• Children's speech is harder for speech recognition than adult speech. A 2022 systematic review of speech recognition for children (Bhardwaj et al., Applied Sciences) notes that most systems are built on adult speech, and that children's speech is challenging because of large differences in how children articulate and sound compared with adults.
• The most promising work on the R sound specifically comes from research classifiers trained on one sound, such as the PERCEPT-R project. In a 2024 study (Benway et al., JASA Express Letters), the best model agreed with human listeners on 88% of correct R sounds but only 70% of incorrect ones. In other words, even a specialist R classifier is weaker at catching the error than at confirming a good attempt. A small 2024 clinical study (Benway & Preston, AJSLP) found AI-assisted R practice promising for five older children and teens, but with a clinician leading the sessions.
• A widely used public dataset for pronunciation scoring, speechocean762 (Zhang et al., 2021), is made up of non-native English speakers (Mandarin speakers, many of them children), not children with speech sound disorders. A good score on that benchmark tells you an engine hears children's voices reasonably; it does not tell you it can spot a lisp.

The honest summary: published evidence that a consumer app can replace a trained listener for children with speech sound disorders doesn't exist yet.

What the Numbers Look Like From the Inside

We can show you one engine's real numbers, because it's ours. Speech Therapy Magic scores attempts with its own engine, ArticScore, and publishes a benchmark page with its caveats. A few figures from that page are worth every parent knowing, whichever app you choose:

• The test set is 300 recordings from child speakers aged 6–15 in speechocean762, judged by five human listeners. Only 74 of the 5,112 sounds were marked wrong by the humans, so every error figure rests on a small sample.
• At its best setting, roughly two out of three sounds the engine flagged as wrong were actually fine to the human judges (precision of about 0.31).
• On sounds the humans scored correct, the engine still flagged about 6.5% as wrong at the everyday "score under 50" cutoff. That's the number a child feels: they said it right, and the app disagreed.
• Of the handful of distortions in the set, it caught 2 of 7. The page calls distortion detection "close to useless," and says the same limit applies to every general phoneme-scoring system its authors know of.

Ask any vendor for numbers like these. The takeaway isn't "scoring is useless." It's that a score is a prompt to listen again, not a verdict.

The Apps, Compared

App	What it listens for	Sound-by-sound feedback	Platforms	Price (checked October 2026)	Best for
Speech Therapy Magic	Speech-therapy target words (23 consonant sounds, by position)	Yes, with a guess at the substituted sound	Any browser (iPhone, iPad, Android, Chromebook, laptop)	$9.99/mo or $79.99/yr after a 7-day trial, no card	Articulation practice between sessions
Google Read Along	Reading a story aloud	Word-level help, not sound-level	Android (Google Play)	Free, no ads or in-app purchases	Early readers, ages up to 12
Microsoft Reading Coach	Reading passages aloud	Flags words to practice	Web, Windows app	Free	Older readers
ELSA Speak	English pronunciation for learners	Yes	iOS, Android, web	Varies in-app; App Store lists $19.99/mo and annual plans from about $71 to $129.99	Teens and adults learning English
Articulation Station Hive	Nothing automatic: an adult scores	Adult taps correct, incorrect or approximate	Apple devices only	Parent $9.99/mo or $83.99/yr, 14-day trial	Accurate home scoring with a trained ear

A few newer apps, such as SpeechLP, Phonomate and AKIO: Speech, also say in their App Store listings that they listen to children and give sound-level feedback. They're small and new, and we found no published accuracy figures for any of them, so we haven't ranked them. If you try one, use the tips below to judge it.

Speech Therapy Magic

Speech Therapy Magic is the one we build, so weigh this section with that in mind. A child taps the microphone, says a target word like "rabbit," and gets a 0–100 score for the word and for each sound in it, plus a plain-language tip ("pull your tongue back and keep your lips relaxed"). It covers the 24 English consonants and 30 consonant blends kids work on in therapy, in the beginning, middle and end of words, across more than 9,200 practice words. Every attempt is logged by sound, so the progress charts show trends for /r/ and /s/ separately.

It runs in any browser, audio is discarded after scoring unless a parent opts in to keep it, and a strictness setting (gentle, standard, strict) lets you make it less discouraging for a younger child. The plan is $9.99 a month or $79.99 a year after a 7-day trial that doesn't ask for a card, with 5,000 scored attempts a month and 3 child profiles.

Where it falls short is the same place every scoring app falls short: it can't reliably hear distortions, and it will sometimes mark a correct sound wrong. Its own benchmark says so. Use it for volume of practice and for spotting patterns over weeks, not to decide whether your child is "done."

Best for: kids roughly 4 and up who are working on specific sounds and need more practice between sessions.

Google Read Along

Google Read Along is a free reading app from Google with an animated helper that listens as a child reads a story aloud and steps in when they struggle with a word. It's genuinely free, with no ads or in-app purchases, works offline after the first download, and Google says voice data is analyzed on the device rather than sent to its servers. The Play listing says it's for ages up to 12.

It's not a speech-therapy tool. It's checking whether a child read the word, not whether their R or S was produced correctly, so a child with a mild articulation error may sail through. For a young child who's learning to read and who enjoys reading aloud, though, it's an easy, free way to get more talking in. We found it on Android; we couldn't confirm a consumer iPhone version.

Best for: early readers on Android who need reading-aloud practice, at no cost.

Microsoft Reading Coach

Microsoft Reading Coach is free for schools and personal Microsoft accounts, runs on the web and as a Windows app, and is aimed at children around 6 to 11. Microsoft says it processes a child's voice to identify pronunciation errors while reading, then gives an accuracy measure and a list of "words to practice." Microsoft also says the audio is processed on the device and not stored or shared. It needs an internet connection and supports US English only.

Like Read Along, it's a reading tool: it's judging whether words were read correctly, not coaching a specific speech sound.

Best for: school-age readers who stumble on particular words, especially if the school already uses Microsoft tools.

ELSA Speak

ELSA Speak is one of the best-known pronunciation scoring apps in the world, but it's built for people learning English as another language, not for speech therapy. It gives detailed, sound-level feedback on accent and pronunciation, and it's polished. Its App Store listing shows a range of in-app prices (for example $19.99 a month, and annual plans from about $71 to $129.99 depending on the offer).

One important detail: the App Store rates it 4+, but ELSA's terms say users must be 13 or older. So it's an option for a teenager working on an accent, not for a young child with a speech sound disorder.

Best for: teens 13 and up who are learning English and want accent feedback.

Articulation Station (No AI, on Purpose)

Articulation Station Hive from Little Bee Speech belongs on this list because it's the best example of the alternative: an adult does the scoring. A child records an attempt, you play it back, and you tap correct, approximate or incorrect. The app turns those taps into graphs, notes and recordings you can share with your SLP. It covers sounds from isolation all the way to conversation, including vowels and vocalic R.

If your SLP has coached you on what a correct sound looks and sounds like, that adult ear will be more accurate than any engine on this list, especially for distortions. The trade-off is that it needs an adult for every attempt. Hive is Apple-only, with a Parent plan at $9.99 a month or $83.99 a year in the App Store and a 14-day trial. (The older one-off "classic" version was retired in December 2024.)

Best for: families on Apple devices who can sit with their child and want the most accurate scoring possible at home.

How to Use a Pronunciation Score Without Being Misled

1. Listen first, then look. Before you glance at the number, decide for yourself whether the sound was right. When you and the app disagree, trust your ear (and ask your SLP about it).
2. Watch the trend, not the attempt. One low score means very little. A sound that sits low for two weeks while others climb is worth mentioning at the next session.
3. Record in a quiet room, close to the microphone. Background TV, siblings and a device across the room all drag scores down for reasons that have nothing to do with your child.
4. Don't let a child chase the number. If a correct attempt gets a low score, say "that sounded right to me" and move on. Kids lose interest quickly when an app tells them they're wrong after they did it right.
5. Ask the SLP what to listen for. A therapist can tell you exactly what a good attempt sounds like for your child's target. That turns you into a better scorer than any app.

When to Skip the App and See an SLP

Apps are for practice. Deciding whether a child needs therapy is a job for a certified speech-language pathologist. ASHA's guidance is that getting help early is better than waiting, and its milestones for ages 4 to 5 say most children of that age produce most consonants correctly and are understandable in conversation. If your child is hard for strangers to understand, is clearly behind for their age, or is frustrated by their speech, ask for an evaluation. You can find a certified SLP through ASHA ProFind. In the US, children under 3 can be evaluated through your state's early intervention program and older children through your public school district, at no cost.

A good rule of thumb from the research on when sounds develop: a review of US studies covering nearly 19,000 children (Crowe & McLeod, 2020) found that, on average, children produced /s/ correctly by about 4 years 3 months, R by about 5 years 7 months, and the "th" in "thumb" by about 6 years 5 months, with wide variation between children. A five-year-old who says "wabbit" is not automatically behind. A free reference with every sound and its typical age is on Speech Therapy Magic's sound library.

How We Evaluated These

We looked at each vendor's own website and app store listing in October 2026 for price, platforms, ages and what the vendor says its speech technology does, and quoted those claims rather than repeating marketing. We did not run a controlled accuracy test of competitors' apps, and we don't publish accuracy numbers for anyone else's product. For Speech Therapy Magic, the figures come from its published benchmark, which is an internal evaluation on a limited sample, not an independent test.

For more on how speech technology works in general, our explainer on natural language processing covers the basics, and our piece on AI and special needs education looks at why speech recognition has historically struggled with atypical speech. If you're choosing an app rather than learning how scoring works, start with our best AI speech therapy apps for kids or our R sound and lisp app guide. SLPs looking at the clinician side should read our guide to AI tools for speech-language pathologists.

FAQ

Can an app tell if my child has a speech disorder?

No. None of the apps here are diagnostic tools, and none claim to be. Only an evaluation by a speech-language pathologist can tell you that.

How accurate is AI pronunciation scoring for kids?

Less accurate than most marketing implies. Our own engine's benchmark shows roughly two in three "wrong" flags were sounds human listeners judged acceptable, at its best setting, and distortions are mostly missed. Most vendors don't publish comparable numbers.

Can an app detect a lisp?

Not reliably. A frontal lisp (tongue between the teeth) sometimes sounds enough like "th" for a scoring app to notice, but a lateral lisp is a distortion, and distortions are the weakest spot for every scoring engine we know of.

Is it safe to let an app record my child?

Check the privacy policy first. Look for apps where the child doesn't create an account, recordings are discarded or deletable, and there are no ads.

Disclosure: Justin Gluska, who writes for Gold Penguin, also builds Speech Therapy Magic, which appears in this article.