Push-to-talk dictation for Mac
Speak. It's written.
Hold a key and talk the way you talk: Cantonese, English, Mandarin, or all three in one sentence. Let go, and clean, punctuated text appears at your cursor in any app.
Free for 2,000 words a week. Pro free for 14 days, no card needed.
Hold fn, speak, let go.
150 ms
median from letting go of the key to recognised text¹
Fast because it listens while you talk
Tachyn streams your voice while the key is down, so letting go only finishes the last word. There is no upload after you stop and no wait for silence. The app shows your own median, measured on your Mac.
Everything you need to stop typing
Every app
Mail, Slack, Notes, your code editor, a browser form: text goes wherever the cursor is.
Mixed languages
Cantonese, English and Mandarin in the same sentence, recognised as one. You never pick a language first.
Clean, never invented
Fillers go, your own corrections are applied, punctuation goes in. Nothing you did not say is added.
Learns your words
Names and jargon you correct become vocabulary, and you can add your own terms and how they sound.
Written your way
Colloquial written Cantonese or standard written Chinese, Traditional or Simplified, UK or US English.
Command, translate, ask
Select text and say "make it shorter". Dictate in one language and insert another. Ask a question and get the answer typed.
Never loses a word
What you say is saved on your Mac before anything is sent. If the network drops, your text is kept to retry.
Light and native
A native Mac app in the menu bar. No browser engine running in the background.
Built for how Hong Kong talks
Many dictation apps expect one language at a time. Tachyn hears code-switching as normal speech, and writes each language the way it should be written, with the right spacing between Chinese and English.
Recognition also covers dozens of other languages, including Japanese, Korean, French, German and Spanish.
- 廣東話
- English
- 普通話
- 日本語
- 한국어
- Français
- Deutsch
- Español
Your words stay yours
Audio is streamed for recognition and is not stored on our servers. Your transcripts and history stay on your Mac. We keep your account email and how many words you dictate, nothing you said.
- No audio kept on our servers
- History only on your device
- Never used to train models
- Delete your account at any time
Try it on your next email
Download Tachyn, hold fn, and say what you would have typed.
¹ Median of 10 dictations of 0.8 to 1.5 seconds each, from letting go of the key to the final recognised text, with a speech session already open, through Tachyn's servers to Soniox realtime recognition, measured by us on 26 September 2026. It excludes clean-up and insertion. Your time depends on your network, your region and how long you speak.