Echo it until it's yours.
Dutch pronunciation, scored sound by sound.
A Dutch teacher's ear for your app. Speech-to-text tells you the words. mnemEcho knows the sentence and tells you which sounds were right.
Send a Dutch sentence and a recording of someone saying it. Get back a score from 0 to 100, a verdict, and every word and every sound marked, with a short English tip on each sound to work on.
What you get back
A score and a verdict
One number, set against native Dutch speakers, and a verdict your code can act on: excellent, pass, marginal fail or hard fail.
Every sound
Each word, each expected sound in it, whether it was said right, what was heard instead, and a plain English tip, such as "the long ee in zee".
The wrong word caught
A learner who says the right sentence with one word swapped (maart for februari) is told which word, not given a pass.
Fair to every voice
A deep, high, male, female or accented voice is never marked down for how it sounds, only for which sounds it makes. We check this on every model change.
One call
curl -X POST https://api.mnemecho.com/api/v1/score \
-H "Authorization: Bearer $MNEMECHO_KEY" \
-F "text=In februari is het Valentijnsdag." \
-F "[email protected]"{
"score": 91.6,
"verdict": "pass",
"worst_word": "februari",
"words": [
{"word": "februari", "ok": false, "wrong_word": false, "sounds": [
{"target": "eː", "heard": "ɛ", "ok": false, "hint": "the long ee in zee"}
]}
]
}Trimmed. The full response has every word and sound.
Built for
Exam prep apps
Know the speaking is ready before the inburgering or NT2 exam.
Schools and teachers
Spoken homework marked sound by sound, so lesson time goes to teaching.
AI tutors and agents
Give your agent a Dutch teacher's ear. The API is described in llms.txt and OpenAPI.
Voice and dubbing QA
Check that a Dutch voice, synthetic or human, says each line right before release.
Games
Speak-to-play: a verdict fast enough to drive the next move.
Speech data teams
Does this recording match its sentence? A score per clip, a flag when it does not.
Private by default
Your audio and text are scored in memory and dropped. They are never stored and never used to train anything. We keep one usage line per request (time, status, audio length, score) to count your limits. Details in the privacy notice.
The learner app
mnemEcho began as an app that trains people to speak Dutch from memory: hear a native speaker, shadow them, then say the sentence from its English, with feedback on every sound. The API runs on the same scoring. The app is in private testing with a small group; write to [email protected] to join.
Getting a key
The API is in early access: keys are free and given by request while we open self-serve sign up. Tell us what you are building at [email protected]. Plans and prices are on the pricing page.