Messages
Your teamTomorrow's meeting
- What you say
- Let's meet at um, three—no, four tomorrow for the project update, the project update.
- Voiced removes fillers, repetitions and self-corrections
- Let's meet at four tomorrow for the project update.
Voiced removes “um” and the repeated “the project update”, and follows your correction to keep the meeting time at four. You can think out loud; the text keeps your final intended meaning.
Speak naturally
A voice. A careful edit.You speak.
You speak.
It edits with restraint.
Fillers, repeats, a change of mind. Voiced cleans them up. Your meaning stays yours.
Every app gets
its own writing style.
Keep messages casual and emails complete.
Switch apps, and Voiced uses the style you set.
LINEStyle · CasualDinner at that new place? See you at seven — I’ll text when I’m close 😂
MailStyle · FormalSubject: Project update Hello, Please find the latest version attached. Could you review it and let me know if anything needs adjusting? Thank you. Best regards
NotesStyle · BulletedGrocery list □ Milk □ Greens □ Tomato □ Banana □ Onion
These examples show how per-app settings keep messages casual, make emails complete and formal, and turn notes into lists. Voiced applies the writing style you set when you switch apps.
Speak in your own voice.
Let the words stay yours.
More than one voice in Traditional Chinese
Traditional Chinese,
voiced.
Mandarin, Cantonese, Taiwanese—and the English in between.
Spoken your way. Written your way.
Voiced · Language & output
You speak · Cantonese with English
我哋聽日，嗯，review 個 proposal，再 confirm 時間。
我哋聽日 review 個 proposal，再 confirm 時間。Original phrasing preserved
Written as · Colloquial Cantonese
In this example, Voiced removes the filler while keeping colloquial Cantonese and the English words review, proposal and confirm together in the same sentence.
Mix Cantonese and English in the same sentence. Keep colloquial written Cantonese or turn it into standard written Chinese—your choice.
Voiced · Language & output
You speak · Taiwanese
明仔載的代誌，明仔載才閣講。
明仔載的代誌，明仔載才閣講。Original phrasing preserved
Written as · Taiwanese Hokkien
This example writes spoken Taiwanese in Taiwanese Hokkien characters, preserving the original phrasing.
Speak Taiwanese and write it in Taiwanese Hokkien characters, or turn it into standard written Chinese.
Dictate anywhere you can type
Use Voiced in email, messages, notes, browsers, and terminals. There is no app switching or pasting—speak wherever your cursor is.
Your words, in 11 languages
Choose your output language. Get translated text when you finish speaking, without pasting it into another app.
Correct it once. Voiced remembers.
Voiced learns from the words you correct, while remembering names, terminology, and phrases you use often.
Traditional Chinese benchmark
25 prompts, shown in full.
The differences appear when English enters the sentence.In this Traditional Chinese test, Voiced got all 8 code-switching prompts right and polished 9 prompts correctly.
Both were the highest results among the five tools.
June 2026 · 25 Traditional Chinese prompts · 5 tools · 1 speaker
In this test, Voiced got all 8 Chinese–English prompts right and correctly polished 9 of 10 prompts, the highest results among the five tools in both categories. All 25 prompts used Taiwanese character forms, and all 25 used full-width punctuation.
Chinese–English code-switching
8 prompts mixing English into the same sentence.
Prompts with correct content| Voiced | 8 prompts with correct content | |
|---|---|---|
| Typeless | 7 prompts with correct content | |
| Wispr Flow | 5 prompts with correct content | |
| BYOK | 4 prompts with correct content | |
| Apple Dictation | 0 prompts with correct content |
Voiced got all 8 code-switching prompts right.It was the only tool with a perfect result; the others got 7, 5, 4, and 0 prompts right.
Polishing behavior
10 prompts that needed polishing, with all four outcomes shown.
Correctly polished; more is better| Voiced | Polished correctly 9Not polished 1Polished unnecessarily 0Meaning changed 0 | Polished correctly 9/10 |
|---|---|---|
| Typeless | Polished correctly 7Not polished 0Polished unnecessarily 2Meaning changed 1 | Polished correctly 7/10 |
| Wispr Flow | Polished correctly 6Not polished 3Polished unnecessarily 0Meaning changed 1 | Polished correctly 6/10 |
| BYOK | Polished correctly 4Not polished 6Polished unnecessarily 0Meaning changed 0 | Polished correctly 4/10 |
| Apple Dictation | Polished correctly 2Not polished 8Polished unnecessarily 0Meaning changed 0 | Polished correctly 2/10 |
Voiced polished 9 prompts correctly without changing the meaning.It left 1 prompt unpolished.
Chinese-only short phrases
9 prompts · prompts with correct content.
| Voiced | 9 prompts with correct content | |
|---|---|---|
| Typeless | 9 prompts with correct content | |
| Wispr Flow | 9 prompts with correct content | |
| BYOK | 9 prompts with correct content | |
| Apple Dictation | 6 prompts with correct content |
Every tool except Apple Dictation had zero content errors. We do not present this category as a Voiced advantage.
Taiwanese writing conventions
25 prompts · prompts using Taiwanese characters and full-width punctuation.
| Tool | Characters | Punctuation |
|---|---|---|
| Voiced | 25 prompts | 25 prompts |
| Typeless | 23 prompts | 25 prompts |
| Wispr Flow | 18 prompts | 19 prompts |
| BYOK | 20 prompts | 25 prompts |
| Apple Dictation | 25 prompts | 25 prompts |
Characters: no Simplified Chinese or non-Taiwanese variants. Punctuation: no half-width punctuation in Chinese text. A prompt can have both problems, so the columns are not added together.
Tested by Voiced in June 2026. The same 25 Traditional Chinese prompts, five tools, and one speaker were used; the results represent only this test.
Manual scoring: changing the meaning, a proper noun, a number, or a negation counted as a content error. Removing fillers, reordering words, or making a sentence read naturally counted as editing, not a content error.
BYOK means a test setup that connects models independently: SayIt provided the interface, Whisper Large v3 handled transcription, and Gemini 2.5 Flash handled polishing.
You decide how your text is written, which information can be used, and where your speech is processed. Set a writing style for each app, control smart personalization and screen context separately, or choose recognition on your Mac.
We protect your privacy.
No training. No sale.
Your data is not used to train AI models or sold to third parties for advertising.
Zero cloud retention for dictation
Audio, original transcripts and polished text for each cloud dictation request are processed in real time, without persistent cloud storage after completion. Local-mode dictation stays on your Mac and is not sent to the cloud.
Clear data practices
Our Privacy Policy explains what data we use, how long we keep it and the rights you have.
Processing and retentionYour data serves this request only.Where it goes, how it is used, and when it is released
The cloud handles this recognition and polishing request. You decide whether to save the returned history on your Mac, and for how long.
Using the cloud does not mean staying there.
These four steps describe the path a dictation takes when you choose cloud mode.
- Your Mac
Only what this request needs
Audio is sent for recognition. The transcript and context you allow are then sent for polishing.
You choose which context is available - Voiced servers
Audio passes through memory only
The voice gateway does not write audio to storage or logs. It is released when the connection ends.
No cloud recording library - Recognition and polishing services
Processed, not retained
Your data is not used for AI training. Text polishing enforces Zero Data Retention (ZDR), and results are released after they are returned.
Used only to return this result - Back on your Mac
You decide how long to keep it
Transcripts and original recordings are stored locally. Set separate retention periods, keep nothing, or delete them immediately.
Your history stays under your control
Zero retention means more than no training.Zero Data Retention means request content is not kept after processing. Not using data for training is a separate safeguard; both apply.
Local history is not a cloud copy.Returned history stays on your Mac according to your settings. You can use cloud recognition and polishing without keeping your own recordings.
This describes cloud mode. Cloud recognition and polishing require an internet connection; local mode recognizes speech on your Mac and sends no audio. Providers and models may change as quality, privacy, and compliance requirements evolve.
Per-app controlsYou set privacy for every app.Choose which information can help in each app
You can use a different writing style and allow different information to be used in each app.
A work email can use more context while private messages use less. For each app, you can separately exclude smart personalization and screen context from recognition and polishing. Screen context is off by default.
MailData use in this app
Mail keeps the assistance allowed by your choices.The background, domains, and tone you enter still apply.
Try the illustrative controls—they will not change your apps. The switches shown are not product defaults; screen context must first be enabled globally.
These choices control extra context; they do not change the core cloud processing used for recognition and polishing. Background, domains, and tone you enter yourself still apply. Global settings also control smart personalization and learning from corrections; learning requires Pro or an active trial. Read the full privacy policy.
Local modeWhen privacy matters more, use local mode.Recognized on your Mac; send it to the cloud when you want it better
Choose free local recognition when you need to work offline or want audio to stay on your computer. When you want cloud results, manually retranscribe the recording.
More platforms
Keep speaking on every device.
Join the waitlist for Windows, iOS, or Android. Choose every platform you use, and we’ll let you know as each one launches.
About Voiced
A few things to know
Where is Voiced developed?
Voiced is developed in Taiwan by Taiwanese developers and operated by ASWAY LTD, a company registered in the United Kingdom.
How is Voiced different from Wispr Flow or Typeless?
Voiced puts Traditional Chinese first without rewriting away your voice. It separates cleanup from rewriting, lets every app use its own writing style, and supports Mandarin, Cantonese, Taiwanese, Chinese–English and Cantonese–English speech. The benchmark above compares five tools on the same set of Traditional Chinese prompts, including mixed-language speech.
Where does my voice go? Who can access it?
In cloud mode, your audio passes through Voiced's secure voice gateway to a speech recognition provider. Audio and text are processed only as needed to return your result, then released without cloud retention. They are not used for AI training. Providers and models may change to meet quality and compliance needs; see our Trust Center for the current details. In local mode, your speech is recognized on your Mac and is not sent to the cloud.
Does per-app styling mean Voiced reads what's in my apps?
Writing styles, smart personalization and screen context are separate choices. Screen context is off by default. If you enable it, on-screen text can inform text cleanup. You can exclude each app from smart personalization and screen context individually. Background, domain and tone information you enter yourself still applies. Smart personalization and learning from corrections require Pro or an active trial.
Can I use Voiced offline?
Yes. Local mode uses Apple's built-in speech model on this Mac, so recognition needs no connection. It is free forever and uses no cloud allowance. It currently supports one primary language, Mandarin or English, plus your dictionary; it does not handle mixed Chinese and English, Cantonese, Taiwanese, polish, rewrite, translation or Chinese-variant conversion. A dictation made in local mode can be sent to the cloud with "Retranscribe" for a better result.
What about Windows, iOS and Android?
Voiced requires a Mac with Apple silicon (M-series) running macOS 14 or later. Intel Macs are not supported. Join the waitlists for Windows, iOS or Android and we will let you know when your selected platforms are available. You can select more than one.
Does Voiced support Taiwanese Hokkien and Cantonese?
Yes. You can also choose to output Taiwanese Hokkien characters or colloquial written Cantonese. In our tests, accuracy is still lower than for Taiwan Mandarin. Taiwanese Hokkien, Cantonese and mixed-language speech are handled in the cloud; local mode currently supports Mandarin and English.
Support & collaboration
Get in touch
For product questions, bug reports or collaboration enquiries, send us a message.
Spoken your way. Written your way.
Voiced
Your voice, voiced.