Original · Unique Research · 2025-11-10
Editorial note: This is a complete English edition of the historical interview and accompanying commentary. Registered-user totals, the number of apps, technology claims and commercial plans reflect the source’s reporting, not independently verified current results. The scope of the “first” real-time speech-recognition claim is not defined in the source. Relative references such as “this year” and “next year” are retained because no separate interview date is supplied. The App Agent is described as under development, not as a confirmed general release. The source names Microsoft’s cloud services without specifying a particular service product.
Many people begin learning a foreign language with vocabulary books and grammar exercises. When Wei Lihua built HelloTalk, he began with people. Years spent living and working in different countries repeatedly brought him up against a small but painful truth: airplanes and the internet were drawing the world ever closer, yet language still frequently kept people apart. In 2012, he decided to build a different kind of language product—not another vocabulary-memorization tool, but a place where people from different countries could talk naturally. HelloTalk was born from that idea: return language to its original form by using it first and continuing to learn through use.
More than a decade later, HelloTalk has accumulated over 60 million registered users around the world. What they do there is actually quite simple: post updates, open voice rooms, watch livestreams, correct one another, chat by voice, and add each other as friends. The complex part is that nearly every step behind those activities rests on AI.
While many people still confine AI to conceptual reports, HelloTalk has embedded it in the smallest details of conversation. Instant cross-language translation helps users catch the sentences thrown their way. Speech recognition reliably turns voices into text across different accents and network conditions. Grammar correction quietly reshapes awkward expressions into sentences the other person can understand without losing politeness. This is not a showy display of AI virtuosity; it is technology placing a soft cushion beneath every user who is afraid to speak.
Its collaboration with Microsoft cloud brought real-time multilingual speech recognition into voice rooms and livestreams, an achievement whose significance extends far beyond the words “technical first.” In a voice room, the greatest danger is not a shortage of people but an awkward silence. A little more latency, one misrecognized sentence, or feedback arriving a beat late may seem like details, but they directly affect whether someone dares to keep talking. Here, AI shortens the few hundred milliseconds between opening one’s mouth and being understood. For a product manager, that is merely a metric; for a beginner, it is the psychological safety that determines whether they dare remain in the room.
In 2020, HelloTalk used TensorFlow technology to build its own English grammar-correction engine, then went on to develop more than a dozen multilingual AI language-learning apps. In technology news, such an accomplishment can easily be written up as just another new feature. From the founder’s perspective, however, it is more like an extended through line: from helping users start speaking, to helping them sound more like native speakers, to ensuring that an AI practice partner is available at any time. AI is not at center stage, but it is gradually reshaping the entire learning journey.
When asked where the greatest breakthrough in generative AI lies, Wei Lihua gave an unadorned answer: large language models based on the transformer architecture represent a crucial turning point. He immediately followed with an intriguing analogy. End-to-end models based on ChatGPT and a socially driven environment for language and cultural exchange such as HelloTalk resemble two parallel versions of end-to-end learning. A large model consumes countless texts from the internet and learns which reasonable word should follow any given word. HelloTalk, meanwhile, carries countless real-world interpersonal interactions, cultural frictions, and shifts in context.
One is a cold data distribution and the other is warm human conversation, but both are trying to answer the same question: the meaning of language lies not in a sentence being correct, but in its being appropriate to the situation. What truly unlocks a language for someone is often not their Nth textbook, but a real conversation. Imagine the first time you comfort a friend you have never met in person, using halting English and discover that your words genuinely comforted them. AI’s value lies precisely in guiding you toward such real moments rather than leaving you forever inside a workbook.
Many companies talk about balancing technological innovation with commercial execution, only for the discussion to become: make some money first, then decide whether to do a little research. HelloTalk has taken almost the opposite path. It has explicitly designated the next one or two years as a period of investment in AI, placed AI at the heart of corporate strategy, and only then considered the pace of commercialization. The company has begun preliminary attempts to monetize AI features, but it cares more about whether those features can truly keep users longer, help them talk more, and make their learning feel more natural. Without this soul-searching at the heart of the product, commercialization can easily deteriorate into attaching a price tag to every button.
While agentic AI is still dominating screens in capital and technology circles, HelloTalk has begun doing something straightforward in its own app: turning the product into a space with a built-in assistant. This assistant is not a robot teacher standing in front of the user. It is more like a coordinator working quietly behind the conversation—helping users identify expressions that may cross cultural boundaries, suggesting more natural ways to phrase a sentence, and even helping them navigate the delicate balance between politeness and authenticity in cross-border social interaction.
If the past ten years were about people adapting to apps, the next ten may see more and more agents grow inside apps and proactively adapt to people. This is especially true in language-based social interaction. Human emotions are too nuanced and cultural boundaries too ambiguous for any one-size-fits-all, templated intelligence to feel anything but crude. An agent’s value is not to speak for you, but to help you dare to speak, know how to speak, and stop fearing mistakes.
Interestingly, HelloTalk has not built an elaborate-looking market matrix for globalization. Wei Lihua’s judgment is that the core product format does not differ greatly across countries. What truly needs fine-tuning is localized adaptation at the feature level and the posture of the operation behind it. Local laws must be respected and local customs understood, but none of this should be designed by headquarters on a whim. Instead, the company uses a more grounded approach: finding local talent from within the user community and working with them.
This reveals a change in the traditional logic of globalization. Companies expanding overseas once treated local teams as a cost item or a compliance option. Now that AI can handle much of the mechanical localization—interface translation, time-zone handling, and multilingual content adaptation—the real value of people stands out more clearly. They are no longer responsible for translating instructions; they translate culture. In a sense, AI has leveled differences at the tool layer, making human experience and insight the new scarce resources.
When discussing the role of Chinese AI companies in the global market, Wei Lihua is both optimistic and clear-eyed. He is optimistic because companies from China possess extremely strong engineering and operational capabilities, honed over years in the high-pressure proving ground of the Chinese market. Once those capabilities are deployed globally, they can generate enormous momentum. More importantly, AI has sharply reduced the cost of global localization for content and operations: the same team can tell different versions of a story for different markets, using one engine with many forms of expression—work that once required a vast overseas organization.
His clear-eyed view is that competition has never been solely about computing power and R&D spending. To build genuine, lasting influence worldwide, Chinese companies must answer three harder questions. Beyond technology, can you tell a story that people in different cultures are willing to believe? Beyond efficiency, can you give users an emotional reason to stay? Beyond short-term growth, can you establish genuine respect for privacy, ethics, and boundaries? Technology can get you into the arena faster, but only these softer and more difficult capabilities determine whether you can remain there longer.
Returning to HelloTalk itself, this is not a company that dominates technology-media headlines every day. It is more like a system running quietly in the background. When you open the App, you see profile photos, updates, and voice rooms. Outside your view, AI is silently correcting a word order, polishing a phrase, or catching a pronunciation you almost failed to understand. Whether the world has become more closely connected does not always require a grand narrative. Sometimes it requires only a simple question: did someone today dare to say one more sentence in a foreign language that they would not otherwise have spoken?
We love turning visions into slogans: connect languages, connect cultures, and bring the world closer together. After hearing such phrases often enough, it is easy to classify them automatically as corporate messaging. Yet for someone such as Wei Lihua, who has spent years building products, this sentence has a very concrete breakdown. If someone meets a friend on HelloTalk and one day dares to travel alone to another country; if a student slowly develops a genuine love for a language by using imperfect phrases to talk about everyday life instead of only completing exercises; or if a middle-aged person suddenly discovers that they, too, can befriend people from other cultures—then the vision is no longer a line displayed on a website. It is a collection of lives quietly placed on new trajectories.
AI’s true value may be hidden in this quietness. It does not necessarily need to appear in the form of a spectacular launch event or exist under the banner of disruptive technology. Often it simply takes the inconspicuous form of a translation button, a correction prompt, a predictive keyboard suggestion, or a subtitle bar in a livestream, silently moving the world’s boundaries forward by a small degree.
When discussing the global future of Chinese AI companies, perhaps we should spend less time asking what new concept can be forced out next and more time asking whether there can be more products like HelloTalk—products that let ordinary people live a small piece of everyday life in someone else’s language. The texture of an era is ultimately determined not by a few PPT slides at a technology launch, but by countless seemingly ordinary conversations. Passing through languages and crossing cultures, they form the world’s true connective tissue.
Selected Interview Q&A
Q1: Could you briefly introduce yourself and HelloTalk?
Wei Lihua: I am Wei Lihua, the founder of HelloTalk. I have lived and worked in several countries, and I founded HelloTalk in 2012. HelloTalk is now a leading global language-social platform with over 60 million registered users, and we have consistently used AI for translation, speech recognition, grammar correction, and social interaction.
Q2: How did HelloTalk first use AI?
Wei Lihua: From the beginning, we used AI for cross-language communication, including chat translation, speech recognition, and AI grammar correction. That also made us a case study for Microsoft’s cloud AI services. Later, together with Microsoft, we brought real-time multilingual speech recognition into voice rooms and livestreaming scenarios for the first time.
Q3: How did you build the grammar-correction capability you mentioned?
Wei Lihua: In 2020, we independently developed an AI English grammar-correction engine based on big data using the TensorFlow technology of the time, enabling users to receive corrections as they chatted in real conversations.
Q4: Besides the main HelloTalk App, do you have other AI products?
Wei Lihua: Yes. We have also built about fifteen AI apps for foreign-language learning, covering multiple languages and addressing more specialized scenarios in language practice and teaching.
Q5: What was your greatest original motivation when you founded HelloTalk?
Wei Lihua: In one sentence: connect languages, connect cultures, and connect people across countries. We hope the product can make it easier for people in different countries to communicate and bring the world closer together.
Q6: In your view, where has the most important recent breakthrough in generative AI occurred?
Wei Lihua: The most important development is still the end-to-end large language model based on ChatGPT and the transformer architecture. In a sense, the boundaries of its capabilities resemble those of the socially driven environment for real-world language exchange that we have built.
Q7: How do you balance technology and business?
Wei Lihua: Our approach is to pursue technological innovation and commercial implementation in parallel. This year and next are both periods of major AI investment. We have already begun preliminary commercialization, and next year’s focus will be on AI-driven growth and monetization at scale.
Q8: Are you working on new paradigms such as Agentic AI?
Wei Lihua: Yes. We are already developing our own App Agent inside HelloTalk. We hope it will help users communicate better across countries and make the product itself understand users better and become more capable of helping them express themselves.
Q9: Do your strategies differ greatly from country to country when expanding globally?
Wei Lihua: Overall, there are no particularly large differences. We make feature-level localized adaptations in a few countries, but the core logic is consistent.
Q10: How do you address privacy, compliance, and cultural differences across countries and cultures?
Wei Lihua: The most important thing is to respect local laws and customs and to localize thoroughly. We find local talent through our user community to help localize content and operations.
Q11: How do you view the role of Chinese AI companies in the global market?
Wei Lihua: Chinese companies have enormous advantages, with very strong technological and operational capabilities. AI makes global localization of content and operations easier, which is a tremendous opportunity. The key is to seize this window and truly embed products in the daily lives of users around the world.