Pronunciation software for language schools: a buyer's guide

Ask any ESL program director which skill their students most need for real life, and they'll say speaking. Then ask which skill gets the least formal assessment, and — after a pause — they'll say speaking again. This isn't negligence. It's arithmetic.

The problem: speaking assessment doesn't scale

Grading one student's spoken response properly takes a teacher several minutes: listen, re-listen, diagnose the sound errors, write feedback. Multiply by 30 students, by 3 classes, by 2 assignments a week, and you get a workload no school staffs for. So speaking gets assessed twice a term in a rushed oral exam, feedback arrives weeks after the error, and pronunciation — the layer that most needs sound-by-sound diagnosis — gets a holistic letter grade at best.

Writing had this problem and got plagiarism-checkers and rubric tools. Reading and listening got auto-graded quizzes decades ago. Speaking is the last skill still graded by brute teacher-hours — which is exactly why an assessment platform is worth evaluating carefully rather than buying the first demo you see.

What to evaluate in a speaking assessment platform

Six questions separate a real ESL speaking assessment platform from a consumer app with a group-billing page:

1. Does it produce per-student phoneme data?

A pass/fail or a single percentage per recording tells a teacher nothing actionable. You want sound-level diagnosis — this student substitutes /s/ for /θ/ and drops word-final consonants — aggregated per student over time. Be especially wary of platforms built on standard speech-to-text: transcription engines are designed to auto-correct errors into real words, which means they literally cannot see many pronunciation mistakes. Ask the vendor how their recognizer handles a deliberately mispronounced word; the answer is revealing. (Ours is a language-model-free phoneme recognizer that scores raw sounds — the approach is explained in how AI pronunciation scoring works — and it's accent-tolerant by design, so students aren't penalized for a regional accent, only for meaning-changing errors.)

2. Can teachers see it without doing data work?

Phoneme data is useless in a CSV nobody opens. Teachers need a dashboard: who completed the assignment, who's struggling, class-wide weak sounds, individual trend lines — at a glance, between lessons.

3. Does rostering and licensing work at your scale?

Fifteen students can be onboarded by hand. Four hundred cannot. Look for CSV roster import, join codes, seat-based licensing you can true-up mid-year, and a hierarchy that matches your org chart — not a pile of individual consumer accounts on a shared credit card.

4. What happens to student voice recordings?

You're collecting minors' voice data. That makes privacy a contract issue, not a checkbox: where recordings are stored, what they're used for (feedback only, or model training and marketing too?), who can export them, and whether access is audited. If you operate in the US, FERPA readiness matters; in the EU, GDPR. Read our position in private, offline pronunciation practice.

5. Is there an on-premise option?

Some ministries, school districts, and corporate training departments simply cannot send student audio to a third-party cloud. If that's you, the shortlist gets very short — most consumer-derived platforms are cloud-only.

6. What does a seat actually cost?

Consumer apps priced at ~$12/user/month look reasonable until you multiply by 400 students, including the ones who enrolled for one term. Look for institutional seat pricing, and check whether teacher and admin seats are billed.

How sayit for Schools answers those six

sayit started as a pronunciation coach for individual learners — per-word, per-phoneme heatmaps with target IPA and articulation tips, four practice modes, IELTS/TOEFL-style exams. The Schools layer wraps that engine in the workflow a program actually runs on:

• Teacher CMS. Teachers write their own paragraphs — this week's vocabulary, the dialogue from unit 4 — and assign to a whole class in one click, or differentiate per batch. Students practice against the teacher's exact material, not a fixed lesson library. There's also PDF import and AI-generated paragraphs when teachers don't want to write from scratch.
• Automated gradebook. Every assignment fills in as students record: completion rates, average scores, per-student phoneme breakdowns. Export to CSV for your LMS or report cards. The grading labor that made speaking un-assessable simply disappears.
• Guardian progress emails. Automated parent/guardian reports — practice minutes, scores, completion — without teachers writing a single email.
• A real hierarchy. Organization → Center → Teacher → Student, with strict row-level isolation: teachers see only their own students, center admins only their center. This is structural, not a UI filter.
• FERPA-ready controls. Access-audit logs record who viewed which student's data; bulk-export lockdown prevents quiet mass downloads of student records.
• Licensing that fits institutions. Seat licensing, CSV bulk import, join codes for self-enrollment.
• Cloud or on-premise. Run it as SaaS, or deploy fully offline on your own infrastructure — the scoring engine doesn't need the internet. There's also a white-label scoring API if you'd rather embed assessment in your own platform.

For students, the sound-level feedback loop — plus the Magic Wand coach that auto-builds drill plans from each student's weakest phonemes — means homework teaches, not just measures. The full learner-side surface is on the features page.

The honest alternative: ELSA for Schools

The main established alternative is ELSA for Schools, built on ELSA Speak's consumer app (25M+ users, 8,000+ lessons covering all 44 phonemes, CEFR level prediction). It's a polished product with real per-phoneme scoring and a teacher dashboard, and its gamified mobile-first design is genuinely sticky for younger students — if daily-streak engagement inside a fixed lesson library is your priority, evaluate it seriously.

The trade-offs to probe in a demo: content is ELSA's curriculum rather than yours (teachers can't assign their own paragraphs against their own syllabus), its scoring model is oriented toward a native-speaker target rather than accent-tolerant clarity, and there's no on-premise deployment. Our detailed comparison of the underlying apps is in sayit vs ELSA Speak.

How to run the evaluation

1. 1.Pilot with one real class for two weeks — not a vendor demo account. Have the teacher assign actual coursework.
2. 2.Run the auto-correct test. Have a student deliberately say "I sink so" and check whether the platform catches the /θ/ → /s/ substitution or congratulates them.
3. 3.Ask for the data-processing agreement before the pricing call. How a vendor answers privacy questions under a deadline tells you a lot.
4. 4.Price the true year: all student seats including churn, teacher/admin seats, onboarding, and what happens to your data at contract end.

See it with your own roster

The fastest way to evaluate sayit for Schools is a demo against your actual use case — class sizes, curriculum, privacy constraints, cloud or on-premise. Contact us and we'll set it up, including pilot pricing (institutional rates differ from the individual plans). Teachers who want to feel the student experience first can try the learner app free in the browser — no card, two minutes to your first phoneme heatmap.