Live site: https://hsk.quannk.com
A self-study guide to Mandarin for everyday conversation, built on the HSK 3.0 word list, levels 1–5 (4,316 words). It has four parts:
- Words: 20 words a day over 217 days. Each word has tone-coloured pinyin, its meaning, and three example sentences with pinyin and an English translation. In levels 1–2, every example also has audio and a short analysis of what the word does in it. Each day ends with a test.
- Conversations: model dialogues for everyday situations. Each comes with structure labels, a meaning for every word, and audio for every line. There are also pattern templates, a phrase bank and words grouped by topic.
- Glossary: tones, sounds, spelling rules and tone changes, with a pinyin chart and optional listening drills.
- Progress: which words you have opened and how your tests went.
It runs entirely in the browser, with no account and no install, and it works on a phone.
I wanted the format that worked for my GRE Lexicon: a fixed set of words each day, examples I can actually understand, and a test at the end. This time it is for Mandarin and for real conversation rather than an exam. Grammar and pronunciation are there when I need them (Glossary), but the daily work is words in sentences and short dialogues.
-
Open Words. It opens on the first day you have not finished; use the arrows or the day picker to move between days.
-
Tap each of the 20 word tabs. Every word shows:
- its pinyin with tone colours and a speaker button;
- its meaning and senses, its measure word, and the Hán-Việt reading of each character;
- 3 example sentences. Tap any word in a sentence for its meaning, and tap the speaker to hear the sentence.
Under More you'll find common combinations, related words and dictionary links.
-
Listen (levels 1–2) plays the whole day hands-free: each word twice, then its three examples, then the next word.
-
Finish with the 21st tab, Test. It shuffles the day's sentences with the word blanked out.
- Pick from 4 options, or type the pinyin (tones optional).
- Missed sentences come back at the end, and the word tabs turn green or red as you go.
- Switch the pool to "Days 1–N" to review everything so far.
Each conversation lists the course day it fits after and how many of its words you have opened. Turn on Structure to see what each line is doing (greeting, request, softener, refusal…), and use Play all to hear the whole exchange. Patterns shows how common exchanges flow, Phrases gives several ways to say the same thing, and Topic words groups vocabulary by situation.
The tones, initials and finals (with comparisons to Vietnamese sounds), spelling rules, tone sandhi and the neutral tone, plus a pinyin chart where every syllable plays a human recording.
Add these to the address to jump straight to a section:
| Link ending | Opens |
|---|---|
#d05 |
Day 5 of the words |
#d05-test |
Day 5's test |
#talk |
The conversations |
#c012 |
Conversation 12 |
#chart |
The pinyin chart |
#drills |
Listening drills |
Your progress (words opened, test results, conversations read) is saved in your own browser on that device.
- Words: the HSK 3.0 list (levels 1–5), ordered within each level by how often each word occurs in film subtitles (SUBTLEX-CH), as a stand-in for everyday speech. Meanings start from CC-CEDICT.
- Examples, analyses and conversations were written by AI (Claude). A script checks every sentence:
- every word must be an HSK word at or below the day's level;
- every pinyin reading must be valid;
- sentences must stay within a length limit.
- Levels 1–2 (days 1–64) had two independent AI reviewers per day, one for natural Mandarin and one for the accuracy of every explanation. A fix pass followed. Then a different model re-read every finished file: it found 156 more serious and 883 minor problems, and they were fixed.
- Levels 3–5 (days 65–217) and their 60 conversations were written by Claude in a single pass, in the style of the HSK Standard Course, and checked by the same script. Apart from day 65 and the first conversation, no AI reviewer read them, so expect more slips there. These days have pinyin but no sentence recordings and no per-sentence analysis.
- Audio: words and syllables are human recordings. Sentences and conversation lines use a neural voice (Qwen3-TTS, generated locally). Speech recognition then transcribed each clip back: for levels 1–2, 4,474 of 4,627 sentences and lines matched syllable for syllable.
No native speaker has reviewed the content yet. If you spot a mistake, please open an issue.
The guide's content is shared under CC BY-SA 4.0, because it builds on CC-CEDICT and on the audio-cmn recordings. See LICENSE.md for the sources and their notices.
This project would not exist without Claude (Anthropic).