Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Show HN: I built an AI conversation partner to practice speaking languages (apps.apple.com)
60 points by omarisbuilding 10 hours ago | hide | past | favorite | 45 comments
Hi,

I built TalkBits because most language apps focus on vocabulary or exercises, but not actual conversation. The hard part of learning a language is speaking naturally under pressure.

TalkBits lets you have real-time spoken conversations with an AI that acts like a native speaker. You can choose different scenarios (travel, daily life, work, etc.), speak naturally, and the AI responds with natural speech back.

The goal is to make it feel like talking to a real person rather than doing lessons.

Techwise, it uses realtime speech input, transcription, LLM responses, and tts streaming to keep latency low so the conversation feels fluid.

I’m specially interested in feedback about: – Does it feel natural? – Where does the conversation break immersion? – What would make you use this regularly?

Happy to answer technical questions too.

Thanks





I just tried in Portuguese, and it introduces itself as ChatGPT. I was going to ask what can this do that ChatGPT can’t, but no need to answer now I guess.

Did it? It uses an openAI (gpt) model. I should refine the prompt. Thanks for the feedback!

The deeper issue is that your product is one of many chatGPT wrappers with no unique hook, not that users can find out you're one of many chatGPT wrappers with no unique hook.

This is only a problem to other sausage makers who feel the competition cheated.

Real users get won't care so long as it works.


but users do care eventually. they just don’t yet have the vocabulary that “other sausage makers” do. dismissing these concerns implies that outcomes matter and provenance doesn’t, which may be convenient if you’re building wrappers, but it’s corrosive if you care about ecosystems, incentives, and long-term quality when the app is compared to others in the market.

that app is not the only app in the market that focuses on conversation. feedback like this is worth taking seriously rather than waving off.


Cute parable but most rely on child sweatshop labor. Users express "thoughts and prayers" level of care if pressed, but not take up a trend of sewing their shirts to spare kids they will never meet RSI.

Noting competition exists seems focused on the outcome of making money. An obligation that exists in an exploitative economy. Where's your concern for the provenance of such obligations?

Yeah. Every expression of concern is probably something like social desirability bias, appearance of concern while sticking with status quo real effort of zero change. Low effort rhetorical "care".

People express concern about global warming and drive off in their SUV. They're just parroting social script.


Cool app. It takes a lot of courage to share here, so well done. I’d definitely highlight a bit more prominently what languages it supports though. The interface is very “Claude” so I’d be keen to hear about:

- How did you go about developing this?

- Was the entire thing vibe coded or just part of it? No shade either way, just curious.

- How long did it take?

- What were some of the harder hurdles to overcome?

- Given the use of AI, what’s your approach to security of your users data? How did you review any generated code?


Thanks! I just wanted to use it myself as it's always hard to find a native speaker to speak with when learning a language. The app is partly vibe coded. I myself am a software developer and had no problem reviewing the vibe coded code. I think it took me a little more than a month give or take, as i was not consistent with my commitment. I would say the syncing part of the conversation was the challenging part, which is still not perfect to be honest. As for user data, i don't store much of user data apart from the minutes used.

Since I don’t own an iPhone, I can only give you feedback on the landing page itself:

- The font in your “Stop studying. Start speaking.” screenshots is both hard on the eyes and strangely blurry.

- Your ad copy needs an overhaul - it feels clipped and rushed.

> I built TalkBits because most language apps focus on vocabulary or exercises, but not actual conversation

There are MANY language apps which focus on actual conversation. You are in a SUPER competitive space. You need to call-out what makes your app different. In just the last few months alone (just on HN) I've seen many Foreign Language Chat Apps:

SpeakLanguageOnline – Voice-only AI language tutor

https://news.ycombinator.com/item?id=46779716

Malan Chat - Full immersion language learning app for 62 languages

https://news.ycombinator.com/item?id=46768430

EnglishCall - AI that calls you and practices spoken English with you

https://news.ycombinator.com/item?id=46714857

TongueFu - Gamified voice-first app for communication

https://news.ycombinator.com/item?id=46553017

Orratio - Practice spoken English by discussing news articles

https://news.ycombinator.com/item?id=46510414


I really appreciate this comment. So much of marketing is "puffery" or based on what would be best for the app creator for the audience to believe. Sometimes that lines up with reality, sometimes it doesn't.

In this case the marketing copy seems more like creator wishcasting or living in the past. I've fallen into this trap before too, of creating things as if AI was a secret in my basement and not something of intense focus by a plurality of 20 million+ software developers. Rarely does a technology landscape change so fast.

How you're framing it is helpful as well. Instead of saying "this already exists (so you shouldn't do it)", I think it's valuable to highlight what a competitive space this is. The creator will need to think hard about what they can do to differentiate their offering in 2026, both in marketing and functionality.


Univerbal - Language learning with a conversational AI tutor

https://news.ycombinator.com/item?id=39188760


Thank you very much for your feedback. I will work on improving the app store screenshots and their descriptions.


I'm curious how the experience is different from using live voice mode in ChatGPT or Gemini?

I haven’t tried this app yet, but my attempts at speaking to ChatGPT in Spanish were pretty dull. Almost every sentence the model replied with “Claro!” and tended to repeat itself often

To be honest i didn't try them. My purpose was just to provide a nice UI/UX to the AI voice model.

You didn't try them!? If you have the ChaptGPT app it's super simple and worth a try. I talk to it in French a lot. It gets tedious though, because after almost every question it tries to end the conversation with "let me know if I can do anything else to help". I really want it to pretend to be curious and continue the conversation. I've thought that this could be fixed with some better prompting, so I'm excited to try out this app.

I always had trouble getting AIs such as ChatGPT to get out of their assistant mode and give natural and conversational replies. No matter what I put in the system instructions it always responds with the same old walls of text and same old it's x not y etc.

Does this one have a prompt that actually takes care of this problem? Does anyone have some nice prompts that make the ai actually useful as a practice partner?


Crazy that when I search for this in the App Store on my phone, it first shows me results for Talbots, and then when I click "show results for Talkbits" it shows me a bunch of other apps. I scrolled through a bunch but then gave up and copied/pasted the URL. Ridiculous!

I'm surprised that an app like this can be rated are 4+, since any open web browser has to be adult-only. I would think that an app like this, unless it has very secure guardrails.

How does Apple deal with rating apps that tie into LLMs?


I really love this concept for an app. Downloading and getting going was a breeze.

That said, I’m finding some latency issues when I respond. I’ll say something in Spanish, hit the red button, and none of the text I sent appears.

Then I’ll hit the red button again to start talking but before I do the assistant responds to my response and the words I spoke show up

Some of the back and forth is showing up out of order.

Also, error messages are in Spanish, and my Spanish isn’t good enough to read them so I’m not sure what to do

Again, love this concept and would love to have an ai assistant I can have daily conversations with to start sharp

Would possibly be nice to prime the ai before the convo saying “I’d like to talk about x” in English

Keep going! I’ll keep playing with it


I pay for LinguaTalk and I also took a close look at TalkPal… how does your feature set compare to theirs?

Congratulations on shipping!

I’ve been working on a similar idea for similar reasons, but only for personal use.

Initially, I thought that if I had a better platform to chat with (better than say, Duolingo) then this would unlock a big growth area for me where I could get more realistic conversational experience. But, as I’ve been building and experimenting, I realize that during conversations I still fall back mostly on the expressions that I’m comfortable with. So, I’ve been experimenting with different modes that will push me to use more advanced forms of grammar and focus in different areas, and so on. Also, I allow myself to select a level of proficiency and dialect (e.g., B1 Mexican Spanish) so that I can get corrections and suggestions that are more specific to my goals.

I’m curious to know if, as a user of your application, you feel like it’s pushing you into awkward situations that will force you to grow your skills?


Thanks for sharing your app. I've been trying a couple of conversations, but the AI is hearing its own voice and then responding to itself, without letting me speak. I hope I'm not doing something wrong, but if I am, this could be a problem faced by many others.

Did you vibe code the whole thing? Talk thru the stack and your experience/findings

Duolingo does this, but only for a few languages and it needs some improvement.

Thanks for your feedback. Yes it definitely needs improvement.

Looks pretty good, the speech pipeline feels noticeably faster than the general-purpose apps I've used for lang conversations. Dead air kills immersion after all

Thanks for the feedback!

Tangential, but on the topic of learning/practicing language, I was thinking that in India, with such high population density, if I want to practice speaking any language, I can find quite a few people within a short walking distance. Personally, I can speak three languages (default for most Indians), but it is very common to find people who speak 5+ languages. I can also understand many other languages that share similar sounds and commonalities to Hindi, so I can be around Gujarati, Haryanavi, Bengali, Marathi, etc., without getting totally lost. Unfortunately, despite living in the South and making many attempts, I could never pick up any Southern language beyond a few words to get by.

Yesterday, while on a walk with a friend discussing SAP, he stopped to greet someone and spoke in Oriya. When I asked, he said he can speak 5 languages fluently and can get by in another 5 or so.

My daughter needs help with her French; we have a neighbor for that (not an App). I’m at three words—Oui, Bonjour, and Bonsoir.


The speaking practice angle is interesting - have you found that users actually stick with it, or does engagement drop off like most language apps? I've noticed the hard part isn't building the AI conversation, it's creating enough novelty/progression to keep people coming back when they could just chat with native speakers online for free.

> chat with native speakers online for free

Where can you do that?


The other comments about calligraphy, copy, competitors, etc are not useful for getting users. This app is already over-engineered.

Go on TikTok and Instagram Reels, scroll for a week 15 mins a day in the language and travel niches. Don't post until you've done that!

Then post funny, scroll stopping videos. In the comments, and in your bio, mention your app.


It's funny, making an app these days is the easier part. Getting people to use it, any substantial quantity of them, is where it's difficult.

You don't need to warm for a full week

What do you need to do instead?

I would be a potential customer at some point in the near future. So I would like to know:

- what languages can it handle perfectly from A1 to C2?

- Pricing

- Any daily caps? I am guessing I can't talk for 24 hours, can I?

- How long can a conversation be until a new one with a new context starts?


The levels would be great. I’ve tried prompting general purpose LLMs to only use words at particular levels, but they pretty quickly diverge from the prompting guardrails. One might just need to building a custom model trained only on a limited vocabulary.

I am on Android. Could you make the app on this platform too?

How about a webapp that runs in desktop/mobile browser?

Does the AI store & train on my voice?


The best part of learning a language is being able to build connections with real people. It's about being vulnerable — learning to take chance — to embrace being uncomfortable. Somehow the uncertainty and vulnerability with other people ends up being one of the most enjoyable parts of learning a language. Taking that chance and getting to know what it's like to communicate with another person across the unknown.

What makes learning a language so wonderful is being, "Lost in Translation."


My observations:

The calligraphic font is antithetical to the theme of your app. The apps colors and ux suggest playful. The font suggests school marm, which is it?

The app itself doesn’t differentiate itself enough to stand out on first use as unique. What does this provide over other similar platforms? How is this different?

The space you’ve chosen is highly competitive. Most of the big players bear a unique signature from ux down to the syllabus they teach.

I applaud you for sharing with us. Sharing here takes grit. Good fortune with your endeavors mate.


What's different between this and all other 10s of different ones doing the same?

This question maybe formulated too harsh, but it is valuable. There are quite a few similar applications (I think I tried 2 or 3 of them). Some are around for couple of years. What is new / unique in your approach?

The real insight here isn't just building another conversation partner—it's the UX optimization for low latency. Language anxiety is largely about response pressure, and even small delays break immersion. The technical challenge of syncing speech input, transcription, LLM responses, and TTS streaming while avoiding feedback loops is non-trivial. Consider adding conversation analytics: tracking where users hesitate, repeat phrases, or switch to easier constructions. That behavioral data could help personalize difficulty curves better than any competitor.

AI slop commenter account above



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: