Education · checked 2026-10-20Can I vibecode ELSA Speak?
KINDA· weekend project
ELSA's loop is simple to describe: show a phrase, record you, score each sound, point at the ones to fix. Cloud speech services now return phoneme-level scores, so an agent can build a personal drill app in a weekend on Azure's pronunciation assessment, with a language model writing phrases from your own text. What you don't get is ELSA's own recognition and scoring models, built on voice data from speakers with various accents, the 8,000-plus lessons and structured courses, exam preparation and the AI role plays. The build is a practice aid, and its scores will not match ELSA's.
$7.50/moprice
$90you'd save a year
weekendbuild time
—people replaced it
What you need on hand.
DifficultyIntermediate · Some system setup or keys.
TimeA weekend; the Azure portal and browser microphone permission take the first hour
- Claude Code, Codex or CursorAny of them can run the prompt.
- Python 3.12 and ffmpegffmpeg converts the browser recording to the WAV format the speech service expects.
- An Azure AI Speech resourceCreate it in the Azure portal; the free F0 tier includes 5 audio hours a month of real-time speech to text. Check how pronunciation assessment is billed.
- An LLM API keyOptional: only for extra practice phrases.
Before you buildCheck Mimora first: it scores your pronunciation offline and for free. ELSA's own free plan also exists, with limits we couldn't read.
- 1. Install one agent: Claude Code, Codex, Cursor.
- 2. Make an empty folder and open it in the agent.
- 3. Paste the prompt below, press enter, and answer its questions as it builds.
The prompt
Build me a personal English pronunciation coach to replace ELSA Speak. Requirements: - Python 3.12 + FastAPI with a single page of vanilla JS. `python -m coach` serves http://127.0.0.1:4870. The page records from the microphone with MediaRecorder and the server converts each take to 16 kHz mono WAV with ffmpeg. Cap a take at 30 seconds. - Scoring: Azure AI Speech pronunciation assessment (`azure-cognitiveservices-speech`) with AZURE_SPEECH_KEY and AZURE_SPEECH_REGION in .env. Pass the reference text, use phoneme granularity with prosody on, and let settings pick en-US or en-GB. Keep the raw JSON of every attempt. - Phrases: paste any text (an email, a script) and the app splits it into sentences. If an OpenAI-compatible endpoint is set in .env, it can write extra phrases that reuse words I got wrong. Model audio comes from Piper (`piper-tts`) and is cached in ./data/audio/. - Attempt view: colour each word by accuracy, click a word for its phonemes with expected and heard sounds, show accuracy, fluency, completeness and prosody scores, and play my take next to the model audio. - Storage: SQLite (sqlite3 module) at ./data/coach.db with phrases, attempts and a running average per phoneme. Takes stay in ./data/takes/ until I delete them from the page. - Drills: a Weak sounds page lists the 10 phonemes with the lowest average over the last 30 attempts; Drill builds a session of 10 phrases that contain them, least recently seen first. - Progress: minutes practised per day, a 14-day chart and a streak counter. - Free-tier guard: DAILY_ATTEMPTS in .env stops scoring after N attempts a day. - Single user. No accounts, no telemetry. Only the recording and reference text go to Azure. - Out of scope: lessons or courses, exam preparation, spoken role-plays, a phone app. - README: creating the Speech resource and key in the Azure portal (budget half an hour), what is sent to Azure, and that scores are a practice aid, not a test result.
Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.
Then keep going
Next prompts, same folder.
The first prompt builds the core. Run these one at a time in the same folder to grow it into your own ELSA Speak.
Extend the pronunciation coach in this folder. Requirements: - Upload an SRT or VTT file and turn each cue into a phrase, keeping its timing. - A shadowing mode plays the cue text with Piper first, then records me, then scores the take; cues I score under 70 are repeated at the end of the session. - A session summary lists the three hardest cues with their weakest phonemes. - Update the README.
Add a conversation mode to the pronunciation coach in this folder. Requirements: - Pick a scenario from scenarios.yaml (job interview, meeting update, ordering food) and chat with an OpenAI-compatible model from .env that replies in two sentences or fewer. - I answer by voice: transcribe with Azure speech recognition, show the transcript, and score each answer with the existing pronunciation assessment using the transcript as reference. - The model's replies are spoken with Piper; keep each transcript and score in SQLite. - Update the README.
Add a weekly review to the pronunciation coach in this folder. Requirements: - Every Sunday at 18:00 write ./data/reports/YYYY-WW.md with minutes practised, attempts, the five phonemes that improved most and the five that stayed lowest, compared with the previous week. - Include three phrases to retry, picked from the lowest-scoring attempts. - `python -m coach report --week 2026-41` regenerates any week. - Update the README.
What you lose
- ELSA's own speech models and scoring, built for non-native speakers
- The 8,000-plus lessons, career lessons and exam preparation (IELTS, TOEFL and others)
- AI role plays and a conversation coach
- A personalised learning path that picks what to practise next
- The iOS and Android apps
Why people still pay
Learners pay for a path they don't have to design: lessons in order, feedback tuned for non-native speakers and a streak that brings them back. Scoring is the easy part to copy; knowing what to practise next and why is what the subscription sells.
ELSA Speak or your own copy?
Over
Your usage
12h ×$/h
ELSA Speak, 1 yr$90
Your own copy, 1 yr$38running + build session
You keep$53pays for itself in 2 months
Estimate · Azure's free tier covers light practice; beyond it, check the per-hour rate. Pronunciation assessment is listed as billed like speech to text, and the pricing page lists it as an add-on. The build is token usage on your agent plan.
Take your data, then cancel.
1 · Take your data out of ELSA Speak
- 1We found no self-service export of progress or recordings.
- 2Individual recordings can be deleted in the Speech Analyzer; ELSA stores recordings so you can review past attempts and doesn't state a retention period.
- 3Access and portability requests go to privacy@elsanow.io; ELSA verifies your identity before acting.
- 4Note your level and the lessons you finished before you leave, because deleting the account removes history and progress.
2 · Cancel ELSA Speak
- 1Website subscription (bank statement shows ELSASPEAK.COM): open the customer portal link in ELSA's help article, enter your email, open the login link you receive and use the cancel button.
- 2App Store or Google Play subscription: cancel in your Apple or Google subscription settings; ELSA can't do it for you.
- 3Subscriptions activated by a voucher revert to free at expiry and need no cancelling.
- 4Your plan or trial stays active until the end date shown, and you aren't charged again.
- Cancelling doesn't refund past payments. The refund policy covers payments made within 7 days: email support@elsanow.io with your receipt.
- To delete the account, open Profile, Edit profile, Delete account; an active subscription may block it, so cancel first. Deletion removes history, progress and the subscription and can't be undone.
- A plan bought in one store can't be upgraded in another, so keep the same channel if you only change tier.
- ELSA says it occasionally uses anonymised audio to improve its models; ask privacy@elsanow.io if you want your recordings removed.
Checked 2026-10-20 against ELSA Speak's docs: 1, 2, 3, 4, 5, 6, 7
The escape hatch
Don't feel like building it? These are already free.
MimoraOpen sourceMIT local pronunciation trainer: it speaks a phrase, you repeat it and it scores the take word by word, offline after install. English is calibrated; a local LLM can write phrases from your text.★ 24 · active Oct 2026 · one-click install
Windows, macOS, LinuxVisit ↗SpeechlingFreeHosted practice with recorded feedback from coaches; a free tier is available and unlimited coaching is a paid plan, so check the free English limit.they host it
Web, iOS, AndroidVisit ↗Microsoft Reading CoachFreeFree Microsoft practice tool that gives real-time pronunciation feedback as you read aloud; aimed at education accounts and reading practice, not lessons for adult learners.they host it
WebVisit ↗
Windows, macOS, LinuxVisit ↗SpeechlingFreeHosted practice with recorded feedback from coaches; a free tier is available and unlimited coaching is a paid plan, so check the free English limit.they host it
Web, iOS, AndroidVisit ↗Microsoft Reading CoachFreeFree Microsoft practice tool that gives real-time pronunciation feedback as you read aloud; aimed at education accounts and reading practice, not lessons for adult learners.they host it
WebVisit ↗
The numbers
ELSA Speak pricing
| Plan | Monthly | Annual, per month | What you get |
|---|---|---|---|
| Pro | — | $7.50 | Pronunciation Coach, topic lessons and study tools (dictionary and study sets): listen-and-repeat drills with sound-level feedback. |
| Premium | — | $8.33 | Everything in Pro plus the Language Coach, AI role plays, Conversation Coach, Speech Analyzer and courses. |
| Professional | custom | — | Everything in Premium plus coaches for public speaking, interviews, presentations, sales and meetings. |
- Free tier
- Not stated on the pages we read; the homepage lists core features as part of a free plan.
- Billing
- On the website only yearly plans are listed. The App Store list shows other options, including a $19.99 Monthly and 3-month plans from $39.99 to $52, and several 1-year prices from $71 to $129.99. Upgrades happen in the store where you bought, and ELSA says cross-platform upgrade isn't supported. The page has a 'Start your 7 days free trial' button, though the plan data we read lists no trial days: confirm at checkout.
- Hidden costs
- Prices differ between the website, the App Store and Google Play, and ELSA's own help article says a plan bought in one store can't be upgraded in another. Exam preparation sits on Premium.
- Source
- elsaspeak.com
Also on the list
Related apps
A free plan exists; the homepage lists a learning path, AI role plays and instant feedback on it, but its limits aren't on the pages we read.
Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: ELSA's own speech models and scoring, built for non-native speakers, The 8,000-plus lessons, career lessons and exam preparation (IELTS, TOEFL and others).
About $7.50 a month on the Pro, billed yearly plan, $90 a year (checked 2026-10-20).
ELSA's own speech models and scoring, built for non-native speakers; The 8,000-plus lessons, career lessons and exam preparation (IELTS, TOEFL and others); AI role plays and a conversation coach; A personalised learning path that picks what to practise next; The iOS and Android apps. If any of those matter to you, keep paying.
Yes: Mimora (open source), Speechling (free), Microsoft Reading Coach (free).