Inference speed and latency
Translation speed on the Neural Engine and GPU, and time to first result, across A-series and M-series generations.
Support development
Help us test Translate Blue on every Apple device.
Translate Blue runs AI translation models on the device itself, so speed, accuracy and battery life all depend on the hardware. We are raising funds to cover every generation of iPhone, iPad and Mac and a year of measuring and improving the app on real devices.
iPhone × 24iPad × 12Mac × 10
About the project
A simulator cannot tell you how a translation app really behaves.
Translate Blue supports more than 380 languages and aims to translate text, speech, camera images and files on the device as much as possible. On-device translation is faster and keeps your data private, but it also means that differences in hardware turn directly into differences in experience.
The same model can run at very different speeds one chip generation apart. An iPhone with 4 GB of memory and a Mac with 16 GB cannot even load the same size of model. Long interpreting sessions run into thermal throttling and battery drain, and each device's microphones and cameras affect speech and text recognition accuracy.
None of this can be reproduced in the Xcode simulator; it has to be measured on real hardware, one device at a time. Today we test on a small set of devices, which is not enough to guarantee quality across the range people actually use, from older iPhones to the latest Macs.
Your support will go towards test iPhones, iPads, Macs and measurement equipment, a dedicated QA engineer, the optimization work that follows from the measurements, and native-speaker quality evaluation. We will share development reports on what we measured on which device and what we improved as a result.
What we measure on real devices
We measure the following on each device and feed the results into model selection and automatic tuning.
Memory limits
How large a model each device can run without the app being terminated, checked model by model.
Heat and battery
Throttling and battery drain over 30–60 minutes of continuous interpreting, to find settings that stay comfortable.
Voice input and live interpreting
Recognition accuracy and latency with the built-in microphones, AirPods and external microphones.
Camera translation accuracy
How camera quality and lenses affect text recognition on signs, menus and documents.
Screen sizes and interaction
One-handed iPhone use, iPad Split View and Stage Manager, and Mac windows, checked on real screens.
How the funds will be used
The goal is based on estimated costs for the devices and equipment we need, plus a year of measurement, optimization and quality evaluation.
| Item | Qty | What it covers | Estimate |
|---|---|---|---|
| iPhone | 24 | Every size and generation from older 4 GB models to the latest Pro Max: speed, memory limits, heat, camera translation | ¥4,800,000 |
| iPad | 12 | iPad, mini, Air and Pro across generations: large-screen layouts, memory and speed while multitasking | ¥3,000,000 |
| Mac | 10 | M1 to the latest generation, 8 GB to 128 GB of memory: large-model inference and PDF translation | ¥6,000,000 |
| Measurement equipment | 1 set | Power meter, thermal camera, measurement microphone, audio interface, AirPods models and cables | ¥1,200,000 |
| Dedicated QA engineer | 12 months | Running and recording measurements on every device, plus regression testing for each release | ¥6,000,000 |
| Optimization work | 12 months | Per-device model selection, automatic tuning and performance fixes based on the measurements | ¥6,000,000 |
| Native-speaker evaluation | 40 languages | Native speakers review translations and speech recognition in the 40 most-used languages | ¥2,400,000 |
| Test infrastructure | 12 months | Mac minis for automated testing and cloud GPUs for comparison runs | ¥600,000 |
| Total | ¥30,000,000 | ||
Prices vary with the market at the time of purchase. Anything raised beyond the goal will go to additional devices and to testing new models as they are released.
Support options
Pick an amount and pay with PayPal. Any amount you choose is welcome too.
¥1,000
A small thank-you
Helps keep development going.
¥3,000
Cables and accessories
Goes towards test cables and adapters.
¥10,000
Measurement equipment
Goes towards equipment such as power meters and measurement microphones.
¥30,000
Part of a device
Goes towards buying an iPhone, iPad or Mac.
¥100,000
One device
Covers roughly one test device.
Back with PayPal
Amount¥3,000
…
Pay with your PayPal account or a credit or debit card.
Frequently asked questions
Which payment methods can I use?
Payments go through PayPal. Besides a PayPal account, you can pay by credit or debit card from the PayPal window. Your card details are never sent to Translate Blue.
What happens if the goal is not reached?
We will still buy the highest-priority devices with whatever is raised and carry on measuring (an all-in campaign).
Will I get a receipt?
PayPal emails you the transaction details once the payment completes. If you need a separate document, please contact support.
Is this related to buying the app or a subscription?
No. Support on this page is a voluntary contribution to development costs and does not change in-app features or App Store subscriptions. It is not a tax-deductible donation.
How are backers and the total counted?
They update automatically when a PayPal payment completes. Backers are counted per PayPal account, so several payments from the same person count as one backer.