Speech Recognition (Lexical)
Goal: Guess the letters you are attempting to speak. Standard Speech-to-Text engines try to be highly forgiving, meaning they will often pass a flat, toneless attempt because they guess your intent from semantic context.
Category Contrast
The fundamental difference: Recognition vs. Calibration.
Most learners confuse speech recognition with pronunciation practice. Understanding the distinction is essential to escaping pronunciation stagnation.
Tone Calibration (Signal)
Goal: Compare your actual voice pitch shape over time directly to a reference speaker. The algorithm measures the pitch contour in a quiet room, showing you where your pitch flattened or dipped incorrectly.
AI formats the cue, not the score
In tone calibration, AI does not decide whether your pronunciation was correct. That judgment comes mathematically from the sound signal. AI is only utilized to translate those mathematical pitch errors into straightforward correction cues.
Core Doctrines
Three pillars of voice calibration methodology.
Đúng Chưa? is anchored in three strict operational principles designed for adult pronunciation production training.
1. Recording check before tone feedback
If the recording is not clear enough, the app refuses to give fake feedback. When audio passes the check, one tone correction is surfaced — not a dashboard of certainty.
2. Physical Coaching Cues
Translates complex signal discrepancies into straightforward pitch guidance. Instead of telling you "your tones are off by 20Hz", the app prompts you to adjust your vocal trajectory: "Keep the pitch steady — dip lower before the break."
3. Quiet-room practice first
Early builds target a reasonably quiet practice space. Stabilize tones in the lab, then use phrases in real life — we do not claim café or street robustness yet.
Early Access
Join the early access beta for tone practice without fake feedback.
Pre-launch validation with serious learners. Join the waitlist for testing updates.
Voice recording is strictly used to evaluate spoken pitch shapes. Early waitlist iterations will store voice samples for telemetry, algorithm debugging, and reference baseline calibration only with clear consent.
You're on the list!
We'll reach out when early testing opens. Want to help us shape the initial practice deck? Tell us a bit about your struggles: