Same microphone, different job

A friend used Vocalace for a week and called it a dictionary that grades you. He already had ELSA on his phone. Both apps take a recording and return a number. After that, they are built for different jobs.
ELSA is the app people name when they want an AI speaking coach: lessons, speaking practice, and a score for how close the accent is, for learners whose first languages are not the same. That is a real product, and a lot of people want it. We did not set out to ship a thinner copy.
Vocalace is a practice loop for Mandarin speakers whose target is General American. You hear a model recording, you say the word, and the page marks the phoneme that went off. The word comes back on a schedule. The score is a practice signal. It is not an exam result, and we do not tell anyone they will end up sounding native. The bar we actually use is smaller: another person can make out the word, and you can hear what you changed.

One first language, one accent

Mandarin speakers miss a short list of English sounds often enough that the product can be planned around that list.
The vowel in cat stays too closed. think slides toward sink. American r gets replaced by a Chinese r, or by l. very picks up a w. water and city keep a clear t where American English flaps it. The vowel in bird turns into something like "bo-de".
Feedback written for every first language has to stay general at the moment it is given. Ours can name the swap Mandarin speakers actually make. The free notes on course.vocalace.com stay on that list: th, /æ/, American r, the flap.
The dictionary, the model audio, and the score all use General American. There is a short US/UK comparison page. Daily practice does not mix the two accents. Mixing them would mean the recording and the phoneme string could disagree, and the score would be grading you against a target you did not hear.

The curriculum is a dictionary

The dictionary has 68,454 entries. 15,359 of those have been checked, and only the checked ones enter daily practice. A checked entry has a General American transcription, a Chinese gloss, a model recording, and up to three example sentences.
The recording is generated from the transcription we signed off on, so the stress and the phones match the line on the screen. Transcriptions are checked by a model first. If that check fails, a person edits the entry and it goes through review again. After the audio is generated, it is listened to again. Mismatches go to a person. Part of speech, gloss, and examples are in that same pass. An entry joins the daily queue only after that.
New words come out by how common they are. Bands 1–10 follow BNC/COCA frequency. Band 11 is checked words that fell outside those lists. Inside a band, a weight breaks ties. Reviews that are already due come out before any new word, oldest due item first. When nothing is due, the session adds one new word.

The pass line and the phoneme mark

Daily practice, example sentences, and tongue twisters pass at 90. That number is the overall score for the attempt. Accuracy, fluency, and completeness are part of it. Phoneme marks sit beside that score and show which sound to work on, next to a mouth-shape reference. They do not decide the pass. On the page, 小 V names the one place to change.
A word passes once that overall score reaches 90. It comes back the next day. A second 90 marks it mastered. If that confirmation attempt misses, it is due again tomorrow, and the next 90 still counts as mastered. After mastery, the gaps are 21 days, 90 days, then 365. A later miss brings it back tomorrow and, once it passes again, the 21-day gap starts over.
We do not drop a weight or shorten an interval the way a lot of flashcard apps do. Pronunciation does not behave like a vocabulary card you either recalled or blanked on. After an attempt is settled, that word cannot show up again the same day.
The daily goal is 30, 50, 100, or 300 words. Trial accounts stop at 30 a day. Paid accounts can go to 300. A new account gets 7 days. The plan is $5 a month or $50 a year, and it does not renew itself.

A clean word still falls apart in a sentence

Flapping, reduced function words, and linking show up once the word sits next to other words. So a mastered headword later comes back as one of its example sentences, with 90 as the pass line. After the first pass, the sentence waits 90 days and comes back once more. A miss, or a skip, brings it back tomorrow.
Tongue twisters are a separate mode: one random line at a time, scored on the same 90 pass line, no repeat until the round is finished.
Intensive practice is separate again. It opens after the day's word goal is done, looks through attempts, picks the phone most worth another pass, and offers three words that contain it. The pass line there is 95. Generating a set costs 5 points, one set per UTC day. If it cannot find words worth drilling, it does not charge. If the sounds from those attempts are already steady, it says so and skips the set.
Search is there for a specific checked word. That practice spends a point and does not write into the mastery schedule. The queue and the search box are deliberately different.

What we left out

We do not run a lesson tree that tries to cover grammar, small talk, and accent in one subscription. The daily loop is short on purpose: hear the model, say the word, see which phone moved, meet the word again later.
ELSA is the better fit if you want a coached path, speaking tasks, and an app that is not built around Mandarin. Vocalace is the better fit if you already know the word and you want the American phones, on a queue, with the recording and the transcription locked to each other.
The site is www.vocalace.com.