Messaging
Reply on Zalo, Messenger or SMS by voice. The floating bubble types straight into the message field.
Anh ơi, em đến muộn khoảng 15 phút vì kẹt xe trên đường Nguyễn Trãi nhé.
Offline AI · Built for Vietnamese
Speak Vietnamese.
See it typed.
Thì Thầm turns your voice into text, right on your phone. Download a model once, then use it with no connection, no account, and your voice never leaves the device.
- Processed 100% on device
- No account, no ads
- Works in any app
0 bytes
of audio sent to a server
5
model sizes, from ~60 MB
1 tap
to speak, one more to finish
100%
offline once the model is downloaded
Made for busy people
Speak faster than you type. Get more done.
From everyday messages to sudden ideas, Thì Thầm helps the Vietnamese-speaking community capture everything by voice, accurately and privately.
Notes
Capture what you need to remember in seconds, even when your hands are busy.
Chiều nay 3 giờ họp nhóm dự án ở phòng 302, nhớ mang theo laptop và bản kế hoạch quý.
Brainstorming
Say the idea the moment it arrives. On phones with 4 GB of RAM or more, the AI adds punctuation and tidies lists for you.
Ý tưởng cho chiến dịch Tết: một là video ngắn về sum họp, hai là mini game lì xì, ba là ưu đãi cho khách quen.
Email & work
Draft emails, reports and work messages without typing every word.
Chào chị Lan, em gửi chị bản báo giá cập nhật, chị xem giúp em trước thứ Sáu nhé.
To-dos
Read out your shopping list or today's tasks. On phones with 4 GB of RAM or more, you get clean bullet points back.
Mua giúp mẹ 2 kg gạo, một chai nước mắm và ít rau muống.
Social media
Write comments, posts and captions as naturally as you talk.
Cuối tuần này cả nhà về quê thăm ông bà, vui ơi là vui!
Tap, speak, done.
Nothing to learn. It all happens on your phone in seconds.
Tap to speak
Tap the mic or the floating bubble and talk naturally, up to 30 seconds at a time.
On-device AI
The PhoWhisper model turns speech into text on your phone. Nothing is sent anywhere.
Paste or insert
The text is copied for you, or inserted straight into the field you're typing in.
Voice typing in any app
Turn on the floating bubble once, then use your voice in any app: messages, email, notes, social media.
- Tap the bubble to speak, tap again to stop
- Text goes straight into the field you're typing in when you turn on the optional accessibility service
- Password fields are always skipped
Accessibility off? The text is copied so you can paste it anywhere.
AI that understands Vietnamese, from tones to context
Thì Thầm is built for Vietnamese speakers, not a multilingual recognizer that's merely "good enough".
Recognition tuned for Vietnamese
PhoWhisper is a model fine-tuned specifically for Vietnamese, trained on speech from many regional accents.
Automatic cleanup
On phones with 4 GB of RAM or more, a Qwen2.5 model adds punctuation and capitals, fixes misheard words from context and removes filler words.
You say
ờ mai mình họp lúc chín giờ à nhớ mang theo hợp đồng với báo cáo tháng mười
Thì Thầm writes
Mai mình họp lúc 9 giờ, nhớ mang theo hợp đồng và báo cáo tháng 10.
Every launch greets you with a Vietnamese proverb: “Uống nước nhớ nguồn” — when you drink water, remember its source.
Privacy isn't a setting
Your voice is your data. Thì Thầm processes everything on your phone and sends nothing.
No audio sent
Audio stays in memory for recognition. It is never saved to a file or uploaded.
No transcripts stored
Text never goes into a file or database. Close the app and the list is gone.
No account, no tracking
No sign-in, no ads, no analytics, no crash reporting.
Network only for models
The internet is used only to download AI models from Hugging Face, each file verified by SHA-256 before use.
Pick the model that fits your phone
Five PhoWhisper sizes, from Tiny (about 60 MB) to Large (about 1.6 GB). The app checks your RAM and free storage and recommends a size.
- Download once, then use it offline
- Runs on the CPU, no GPU needed
- Interrupted downloads resume automatically
Frequently asked questions
Does Thì Thầm send my voice to the internet?
No. Audio and text are processed on your phone and never sent anywhere.
Do I need an internet connection?
Only to download an AI model the first time or when you switch models. After that the app works fully offline.
Which languages does it recognize?
Vietnamese only. Thì Thầm is optimized for Vietnamese to be as accurate as possible. The app itself is in Vietnamese.
How long can I speak each time?
Up to 30 seconds per recording. If there is no speech in the first 5 seconds, recording stops by itself.
Will it run on my phone?
It runs on the CPU, with models from about 60 MB. Automatic punctuation needs a phone with 4 GB of RAM or more.
When does it launch?
Thì Thầm is coming soon to Google Play and the App Store. Email us and we'll tell you when it's out.
Ready to talk instead of type?
Thì Thầm is launching soon. Leave us your email and we'll let you know the moment it's available.
Google PlayComing soon
App StoreComing soon
Get notified: dev@itgratevn.com