中英混说 is the normal case, not the edge case
Say the sentence the way you would actually say it: “这个 feature 下周才能 ship，先让 Michele review 一下.” Most dictation tools panic here — English words become random characters, or the whole sentence gets forced into one language. Alder routes Chinese to an on-device engine we specifically measured on code-switching audio, because monolingual test scores tell you nothing about how a tool behaves when languages switch mid-sentence.
The result: your English stays English, your 中文 stays 中文, and the sentence reads like you wrote it — not like a machine translated half of it. When we ran mixed Chinese–English recordings through the candidates, the difference wasn’t subtle: on the same audio, the specialized route made dozens of times fewer errors than the universal engine.
Details Chinese users notice immediately
- Native punctuation. ，。？「」 — not English commas and periods scattered through a Chinese sentence.
- One script, consistently. Choose 简体 or 繁體 and the output stays there. No drifting between scripts in the middle of a paragraph.
- Your names, your spelling. Personal vocabulary keeps product names, colleagues, and brand terms exactly the way you write them.
Cleaned up, or exactly as spoken — your call
Cleanup in Alder is a switch, not a personality. Turn it on and “嗯就是那个…我们我们下周 发布” becomes “我们下周发布” — fillers and stumbles gone, meaning untouched. Turn it off and Alder types every word exactly as you said it. Some days you want polish; some days you want a verbatim record. Both are one click away, and nothing rewrites your meaning either way.
Speak Chinese, send English
Alder translates on the device too. Set the output language to English, dictate in 中文, and what lands at your cursor is a finished English sentence — useful when you think faster in Chinese but the email needs to go out in English. It works in the other direction as well, and “Same as Spoken” keeps translation off entirely until you ask for it.
If something goes wrong, you’ll know
A Bluetooth headset that connects but never sends audio. A recording interrupted halfway. A text field that refuses input. These things happen, and Alder’s rule is simple: it never pretends. If no usable audio arrived, it tells you instead of typing nothing and calling it done. If you quit mid-dictation, the audio is saved and offered back on next launch. You spoke it — you keep it.
And because everything runs locally, none of this ever involves a server: 你说的话不出这台 Mac。
常见问题 · FAQ
- Does Alder handle mixed Chinese and English? 支持中英混说吗？ Yes — mixed speech is the main case we test for Chinese. Say "这个 feature 明天 ship 不了" and the English words come out as English, in the right place, without the engine guessing them into Chinese.
- Will the output use Chinese punctuation? 输出是中文标点吗？ Yes. Chinese dictation gets native punctuation — ，。？「」 — not English commas dropped into Chinese sentences. Output also stays in one script: pick Simplified or Traditional and it will not drift between the two.
- Is my voice uploaded anywhere? 语音会上传吗？ No. Recognition, cleanup, and translation all run on your Mac. Dictation audio and transcript text are not uploaded — there is no server on the other end to send them to.
Alder is built the same way for every language it supports — Japanese dictation · 日本語音声入力, Korean dictation · 한국어 받아쓰기, German dictation · Diktieren auf Deutsch and Spanish dictation · Dictado en español run on the same on-device pipeline. The engineering behind it is in Why Alder runs multiple speech models.