Answers to the questions learners ask most

SoraSpeak is a Japanese speaking-practice platform. You hold a spoken conversation with an AI teacher, and every turn comes back scored for pronunciation, grammar and fluency. Corrections are written in English, so you can read them without already being fluent in Japanese. It is built for learners who can read Japanese but freeze when someone speaks to them.

The JLPT has no speaking section. It is possible to pass N2 without ever having held a conversation, and many people do. SoraSpeak is the practice that gap leaves behind.

Yes. The free tier requires no credit card and includes 3 sessions per day with Aoi, 2 topic categories, and a 7-day session history. You are never charged unless you choose to upgrade.

Any modern browser with a microphone — phones, laptops, tablets, desktops. There is no separate app to install. The web app works on iOS, Android, Windows and macOS browsers.

No. The microphone built into your phone, laptop or any standard headset works. A reasonably quiet room helps the speech recognition hear you clearly.

No. Speech recognition and the AI conversation both need a server connection. A typical home or office connection is enough.

Kana helps, but you can start without it. What the teacher says is shown as Japanese text alongside a live translation in your own language, so you can follow a conversation before you can read one fluently. Corrections are always explained in English.

The interface is available in 19 languages, including English, Vietnamese, Japanese, Korean, Chinese, Indonesian, Thai, Spanish, Portuguese, Hindi, Turkish, French, German, Arabic, Russian and Italian. The practice conversation itself is always in Japanese.

Because in Japanese the distinction that changes how you speak is politeness register, not regional accent. Aoi teaches everyday 日常会話 — the plain and polite forms you use with friends, in shops and on trains. Ren teaches business Japanese and 敬語 — the honorific and humble forms you need at work. A third voice teaching the same register would be a choice with no consequence.

Aoi, unless you specifically need Japanese for work. She speaks more slowly, waits longer before prompting, and rephrases rather than repeating when you stall.

Yes, at any time. Many learners run daily sessions with Aoi and switch to Ren before an interview or a meeting.

Yes. Ren corrects the politeness level itself — a plain form where 敬語 is expected, or over-formalising a casual exchange. Aoi concentrates on keeping you understood and the conversation moving.

Scoring uses a production-grade speech recognition engine for transcription plus a separate scoring pass, and handles non-native Japanese pronunciation well. Treat the score as a directional signal for which sounds are giving you trouble, not as a substitute for a human examiner.

Scoring concentrates on whether you are intelligible and on the sounds learners most often get wrong. Pitch accent varies by region and by speaker even among native speakers, so it is not treated as a pass/fail dimension.

A production-grade multilingual ASR engine on a low-latency inference platform, configured for Japanese. Specific vendor and model are not disclosed for operational reasons.

A purpose-tuned conversational language model, with a per-teacher system prompt encoding persona, politeness register, difficulty band and teaching strategy. Specific vendor and model are not disclosed.

Pronunciation scoring is grounded in analysis of your actual audio, which limits invention on that side. The conversational reply is AI-generated and can occasionally produce phrasing that is grammatical but unusual. Where it matters, compare your audio against the native sample provided.

Consistency matters more than session length: short daily sessions beat one long session a week. Individual results vary widely with starting level.

For daily speaking volume, an AI teacher is available at any hour, never makes you feel embarrassed for freezing, and costs a fraction of a tutor. For nuanced cultural coaching and long-term motivation a human still adds something AI does not. Most learners benefit from both.

Those build recognition — you see or hear something and identify it. SoraSpeak builds production: you have to say it out loud, unprompted, in a conversation that reacts to what you actually said. The two are complementary, and recognition alone is precisely why people pass reading exams and still cannot speak.

A general voice assistant will talk to you in Japanese. It does not score your pronunciation, does not hold a consistent politeness register, and does not track progress between sessions. If your goal is measured improvement rather than exposure, a tool built for it is what supplies the measurement.

Marketplaces connect you with human tutors, usually priced per lesson or per minute. SoraSpeak is AI, has a free tier, and is available at 2am. For nuanced feedback from a person who knows you, a human tutor still wins. For daily repetition without watching the clock, the cost difference is the whole argument.

No, deliberately. The JLPT tests reading, listening and grammar, and there are good tools for those already. SoraSpeak covers the one thing the JLPT does not test at all.

Audio is processed for transcription and scoring. Sessions are recorded so you can listen back during the session, and Premium users can review past recordings as part of full history. Audio is never used for advertising or sold to third parties.

Profile data, session history and stored audio references are deleted. Aggregated, non-identifiable analytics may be retained for product improvement.

Accounts and session metadata are stored on managed PostgreSQL infrastructure. Audio is processed by a production-grade ASR engine for transcription. Your data is not transferred outside these processors for marketing.

Yes. Request an export or deletion through the support link at the top of this page at any time. Data-handling practices are documented in the privacy policy at /privacy.

You wait until the next day for it to refresh, or upgrade for unlimited sessions. There is no credit card on the free tier, so you are never charged unexpectedly.

Premium is $12 per month, $30 for 3 months, or $79 per year ($6.58/month equivalent — 45% less than paying monthly). It unlocks unlimited sessions, both teachers, all 320 topics across 10 categories, Speaking Challenges, Role Plays, full session history, detailed analytics and a per-sound error breakdown. Lifetime access is available by email.

Cancel from the billing page in your account settings. The subscription stays active until the end of the current period, then reverts to the free tier with no further charges.

Discount eligibility is reviewed on request. SoraSpeak was built by students — use the support link at the top of this page and include proof of enrolment.

Refunds within 7 days of subscribing are typically honoured if you have not used Premium features extensively. Contact support to request one.

The browser must have microphone permission for this site. Check the microphone icon in the address bar, or your operating system's privacy settings, then reload the page.

On iPhone and iPad, check the hardware silent switch first — it mutes some kinds of web audio without showing any error. Then tap the on-screen control once: iOS requires a real tap before a page is allowed to play sound.

Usual causes: background noise, the microphone sitting too far from your mouth, or speaking quickly without natural pauses. A headset in a quiet room typically moves the score more than any change in how you speak.

The avatars need WebGL. Try the latest Chrome, Edge or Firefox, confirm hardware acceleration is enabled in browser settings, or update your graphics drivers.

AI conversation is mostly good but occasionally drifts. Restart the session — each one is independent, so nothing earlier is lost. If a specific problem keeps recurring, send the session details to support.