The microphone is managed by the browser itself: the child pronounces the word, it fills in letter by letter, and none of their audio is saved anywhere.
In Zuki Trace and Read's Reading mode, the word appears hollow; the child says it aloud, and the letters light up as they are understood. There is no «record» button because nothing is recorded.
When a reading session begins, Zuki asks for permission to listen once and opens the microphone using the browser's Web Speech API. The recognition is performed by the device's own browser, in the profile's language (es-ES, en-US, fr-FR, de-DE, it-IT, or pt-PT): the app does not record audio, does not save any sound files, and only the word's outcome (correct, skipped, and in how many attempts) reaches our servers. The waves that move while speaking read the microphone's instantaneous volume and store nothing.
The word is displayed with hollow-outlined letters and fills in white, letter by letter, as soon as the recognizer begins to understand the start of the word. The fill only advances, never recedes, and digraphs light up together (in Spanish ch, ll, rr, qu, and gu) because they sound as a single vocalization. It's a reading karaoke: the child sees the progress as they speak, not afterwards.
Comparing a child's voice with a recognizer made for adults often fails, so the comparison is deliberately generous. The word is normalized without accents, passed through a phonetic table specific to each language (in Spanish, ll sounds like y, v like b, z and c in ce/ci like s, h is silent, and final -d and -s are forgiven), and measured with Levenshtein distance: 1 error is allowed for up to 4 letters, 2 for up to 7, and 3 for longer words, one less error in «difficult» mode and one more in «easy» mode. It also counts if the word is said within a phrase («es una casa») and the four alternatives returned by the recognizer are checked. We prefer a generous correct answer over telling a four-year-old «wrong».
If they get stuck —after three attempts or between 8 and 18 seconds depending on the configured voice speed—, Zuki pauses the microphone, breaks the word into syllables using real separation rules, and reads them slowly, highlighting each one; then it listens again. The child can tap any syllable to hear it again, press «Listen» for Zuki to say the complete word, or pause the microphone whenever they want. Attempts are unlimited and do not deduct points.
Their voice is not recorded
the microphone is opened and processed by the device's browser using the Web Speech API. The app does not create any audio files, and only whether the word was correct and in how many attempts reaches our servers.
Letter-by-letter filling in real-time
hollow letters light up as you speak, not when you finish. Progress only advances, never recedes, so you don’t see what you’ve already said correctly 'turn off'.
Tolerance designed for children’s voices
phonetic table per language plus Levenshtein distance with a margin based on word length (1 error up to 4 letters, 2 up to 7, 3 for long words) and according to profile difficulty.
Never tells you that you’ve made a mistake
there’s no error screen or red-counted attempts. If it doesn’t work, the word remains there and you can try as many times as you want.
Automatic syllable-based help
After three attempts (or 8-18 seconds depending on voice speed), Zuki pauses the mic, separates the word into syllables, and reads them slowly, highlighting each one. Then it listens to you again.
Six languages with their own rules
the recognizer starts in the profile’s language, and each language has its phonetic table and its digraphs (ch, ll, rr, qu, gu in Spanish; ch, sh, th, ph, ck in English; sch, ei, ie in German…).
Manual mode if there’s no microphone
if the browser doesn’t support voice recognition, permission is denied, or the recognizer fails three times in a row, the session continues with the «I said it!» button and the same celebration. Nobody is left staring at an unresponsive screen.
The adult controls everything
permission is requested only once with Zuki's «May I?», the microphone pauses with a button, and in the adult zone, the comparison difficulty and Zuki's speaking speed are adjusted.

Create their profile, choose Reading mode, and with the first word, you’ll see the letters light up as they speak.