TransHero is not a single translation engine. It is "conversation infrastructure" that integrates 12 features, each designed around the lived realities of foreign residents in Japan. This chapter introduces five designs where the differences from other services are most pronounced.
Feature 02 — Voice Engine
Voice Slicing × Multi-AI Coordination — Robust in Noise, Stable Over Time
TransHero's simultaneous-interpretation engine slices each utterance into millisecond-level voice fragments and runs a deliberative architecture: multiple specialized AIs operate in parallel — one for speech recognition (ASR), another for code-switching, another for scene-context inference, another for honorific-tier selection, another for translation — with their outputs voted on and reconciled.
This stack is the product of several years of patient build-up. We observed and interviewed in the field the situations foreign residents actually face, then systematically collected, surveyed, and organized the patterns of mistranslation and misrecognition that occur there, classifying the specialized vocabulary, conventional expressions, and honorific tiers of administration, healthcare, education, housing, and business. Licensed commercial foundation models sit at the base; on top of them we layer the field data we have gathered ourselves, then run each AI through dozens of fine-tuning iterations tuned for its specific scene. The voice-fragment deliberation scheme itself was refined by closing one real-world failure at a time.
What we have assembled is a stack of our own design — and it is what makes the two capabilities the field demands, noise resistance and long-context retention, hold up at the same time.
Anti-noise
Noise Resistance — Holds Up in Hospital and City-Hall Environments
Multiple AIs vote on and reconcile the recognition output of each slice. The result: stable transcription even amid the environmental sounds — paging announcements, voices from the next counter, coughing, paper rustling — that a single model would miss or misrecognize. Many of the generic-ASR errors observed in our empirical validation (e.g., "My Number" → "number") are avoided by this deliberative approach.
Long context
Long-Context Retention — Translations Stay Consistent as the Conversation Builds
While translating slice by slice in real time, a separate AI continuously tracks the session's overall context — opening framing, proper nouns, honorific relationships, scene tag. As a result, even in a long simultaneous-interpretation session, "the names and procedure terms introduced early on" do not drift later. Consultations, interviews, business negotiations — anywhere context accumulation drives the outcome — are where this design shows its edge.
A deliberative structure where voice fragments are voted on and reconciled by multiple specialized AIs. Licensed commercial foundation models sit at the base; on top of them, field data we have observed and collected ourselves, specific to the lives of foreign residents in Japan, drives continuous retraining and retuning. The result is a stack of our own design — engineered by closing real-world failure patterns one at a time, the patterns that generic translation handles worst: real-world noise and long-form context.
Feature 03 — Domain Specificity
Scene Tags × 6 Categories
When the user selects a scene tag, TransHero invokes the corresponding official glossary and honorific tier. This is at the heart of the differences observed in the empirical validation in this report.
市役所
Administrative procedures: residence records, national health insurance, taxes
入管局
Statutory procedures: residence-period extension, status changes
銀行
Financial procedures: account opening, remittance, home loans
病院
Medical scenes: intake, symptom description, prescription explanation
学校
Communication between guardian and teacher; renraku-cho register
一般
Everyday conversation; general interpersonal communication
Feature 05 — Business Model
Hero Points — An Answer to "Subscription Fatigue"
TransHero does not use a monthly subscription; it adopts a pay-as-you-go model where 1 second = 1 point. Basic features (text translation, photo translation, TTS read-aloud, offline phrasebook) are fully free; only extended simultaneous interpretation consumes points.
Earn
Earn in the Community
Earn translation minutes through real-name verification, posts, comments, and purchases on StayHero. Even without spending money, contributing to the community covers everyday translation needs.
Buy
Buy What You Need
Top up via activation code. Three packages: 30, 100, and 300 minutes. Unused points remain valid long-term — nothing wasted.
Free Forever
Basic Features Always Free
Text translation, photo translation, TTS read-aloud, and the offline phrasebook are permanently free. "Translation when you really need it" is available without a subscription.
For many foreign residents, even a subscription of a few hundred yen per month — "a flat fee for something I might not use" — is a psychological burden. Hero Points brings a community-style economy to translation: pay for what you use, contribute and use it free.
Together, these 12 features make TransHero "conversation infrastructure" rather than a translation engine. The differences in translation quality shown in this report should be understood not as a single-feature gap, but as a difference in overall product-design direction.
TransHero