I recently realized BunPro allows for example sentences to be read out load. I assume the recordings are generated via TTS. If that’s the case, what’s the service being used?
3 Likes
Nevermind, it says: “Digital media voiced and recorded by Haruna Oka”. That’s why it sounded so impressively natural…! 
9 Likes
They actually use rabbits to do their voices. It’s quite impressive.
12 Likes
Those are the dulcet tones of BunPro’s very own Haruna san. As I’ve heard it a lot only in the context of doing BP reviews, I think I’d find it a very weird experience to ever speak to her to person, a bit like meeting a celebrity I guess. I had the same when I met a colleague with whom I zoomed with for 2 years before we met in person.
5 Likes
Actually “only” the grammar lesson’s sentences are human voiced.
By the way I’m wondering if the male voice actor is a native speaker. It sometimes sounds weird to my beginner’s ears.
Exemple sentences for vocab decks are TTS and still a bit off now and then, but it’s a work in progress.
4 Likes
It’s a program called Haruna, a very advanced AI, currently not available to the public, developed in Japan. Very few people can see it in action, and it probably comes with a very expensive monthly fee. They also use it to generate some practice content here in the forum for free!
You can also interact with her language model by invoking @ Haru, as of late, there’s no increased subscription cost for doing so. 
7 Likes
Haruna does most of the audio for grammar! However we are looking more into AI in the future, as it is starting to get more and more ridiculously real. A good example is Elevenlabs. If you put Japanese text into some of those voices it’s like ‘whoa, this isn’t a person??’ 
I completely understand the ‘but it’s AI’ argument, but once it’s perfect, it’s perfect.
4 Likes
Not sure if I enjoy Eleven Labs. As someone working with GenAI for over a year and using many TTS products/services targeting the Japanese language, I find all of them very tricky to get right or natural sounding.
If you’re referring to the tts on bunpro, I wouldn’t worry about it. They’re replacing all tts with native audio from professional voice actors 
2 Likes
Not to worry, in 20 years TTS will define what natural language is supposed to sound like, which is finally going to reconcile this issue.
New generation will mimic language generated by AI idols, AI live streamers, AI lessons/lectures and AI news broadcasts.
Language used by us aging meat bags will sound to them as quaint as お主、何を言っておるのじゃ sounds to us now.
Maybe.
1 Like
I believe the TTS they use for vocab is VoicePeak however I may be wrong, or they might have changed what they use.
Yes!!! Very impressive!!! 