Currently I am at around a N4 level. Speaking Japanese fluently is high on my list of Japanese goals. I spend 30-60 minutes a day driving on my own. The departure times vary. To utilise this time I have been trying out AI conversation partners, for voice only conversations. I could write 10000 words on this subject but will be concise for the sake of brevity. I have tried ChatGPT Plus (5.2 currently) as well as the other major offerings such as Grok and Claude and Gemini. I’ve also tried many apps such as LangoTalk, LanguaTalk, Speak, and a few others.
I could summarise my experiences by saying they all suck tremendously and none of them sound remotely like any Japanese person I’ve ever spoken to in my life. The problem is likely the training data; they all speak like they’re reading out loud a written response based on written Japanese scraped from the internet which is precisely what they are doing.
I can also say it’s been excellent. Why? Because regardless of what the LLM is outputting, I am forced to come up with a response in Japanese. It’s great practice and there’s zero embarrassment or awkwardness. That’s been really handy.
What is really shocking is that none of the apps readily tell you what model and what training data the are using. You’d think this would be basic info and easily accessible but no. So when I tried all those apps I didn’t ultimately know if they were all just repackaged versions of ChatGPT which is what I suspect anyway.
Furthermore I am aware of many other Japanese first LLMs that have demos online. The best I’ve found is J-Moshi. The model sounds insanely good. I’m totally not a programming savvy person though. If there is a way to use such a model please enlighten me but from what I can see it’s just a demo.
Anyway it’s Jan 2026 and LLMs are rampant but there seems to be this huge chasm where nobody has managed to make a spoken Japanese voice chat model that’s commercially available and would suit a beginner Japanese learner, and that can be set at different levels of complexity, vocabulary and grammar. Or maybe there is and I just don’t know about it or how to access it.
For now I will keep using Langotalk. I do not recommend that app to anyone. It’s probably worse than shadowing using YouTube videos or beginner level podcasts. But for me it forces me to say things in Japanese and I find that helpful.
There are many threads on the forum about varying use of LLMs but I didn’t see much about voice mode LLMs and the plethora of “AI Japanese tutor” apps that are flooding the iOS App Store, more and more every time I look.
Who’s tried Langotalk, PingoAI, or the main offerings from OpenAI et al in voice mode while driving around? Does anyone know how to use some of the made-for-Japanese people models like J-Moshi? Does anyone have any helpful pointers for me?