He left MIT before it gave him a degree, helped Google build a face computer nobody wanted, then decided the next interface wouldn't have a screen at all - and started a Shenzhen company that recognizes you in three seconds of speech.
Lawrence Chen runs an AI company in Shenzhen called SpeakIn, and the work he is doing this year is the same work he has been doing since 2015: convincing institutions that the most reliable thing about a human being is not a password, not a face under bad lighting, not a fingerprint left on a glass - but a voice. SpeakIn's engine, which the company says hits 99.5 percent accuracy against an industry baseline of 95, sits inside Chinese banks helping verify remote customers, and inside public security and judicial systems helping investigators tell one voice from another in a database. A voice goes in. An identity comes out. That is the whole business.
SpeakIn sells four things in one box: voiceprint recognition (it is you), comparison (these two clips are the same person), verification (the person on the line is the account holder), and liveness detection (you are speaking now - this is not a recording somebody held up to the microphone). The liveness piece is the one Chen talks about most, because it is the one the rest of the industry keeps getting wrong. Synthetic voices have gotten very good. Recordings have always been good. A useful voiceprint system has to know the difference, in real time, with no second take.
Chen's pitch to the financial sector is the simplest version of the long argument: passwords are a temporary inconvenience the world has agreed to endure. They will be replaced. They are already being replaced. The replacement, in his telling, is biometric - and of the biometric options, voice is the only one that works through a phone you already own, in three seconds, while you do something else.
The argument has a backstory, and the backstory is Google Glass. Before SpeakIn, Chen worked on the human-machine interaction program for Glass - the famous face computer that famously did not become a product. He has called the experience formative. The lesson he took from it was not about wearables. It was about screens. Specifically: when the screen gets small enough, or disappears entirely, the user loses the visual feedback that keeps every other interaction model working. You can't tap what you can't see. Voice is what is left.
"Under these preconditions," he told a reporter in 2017, "voiceprint will be the best choice for logging into these smart gadgets with small screens or without screens." It is the kind of sentence that sounds obvious now and sounded like a bet then. He made it before voice assistants were everywhere. He acted on it before anyone called this category a category.
He went to MIT for applied mathematics and engineering. He did not graduate. There is a version of the founder story that treats this as the headline - dropout becomes CEO, etcetera - and Chen, to his credit, does not lean on it. He treats the MIT years as research and the Google Glass years as fieldwork and the SpeakIn years as the part where the answer finally compiles. The team he built around himself, after he came back to China, has degrees from MIT, Harvard, the Hong Kong University of Science and Technology, and Microsoft Research Asia. Over 60 percent of SpeakIn is in R&D. This is a founder who hires engineers and treats sales as a downstream consequence of being right about the technology.
SpeakIn is headquartered in Shenzhen, the city most likely to ship a new piece of hardware on a deadline you would not believe. But Chen kept an R&D arm in the United States from day one. The split is unusual and deliberate. Shenzhen is where his customers are - the banks, the public security bureaus, the device makers who want voice-first login. The US arm is where some of the algorithmic work happens. The company runs its training on Nvidia GTX 1080 cards and Xeon E5-2697A v4 processors, and its product surface mentions a roll call of signal processing primitives that would make a DSP textbook blush: Hamming, Hann, Blackman-Harris, Kaiser, Bartlett, Welch, Gaussian, Blackman, rectangular. These are window functions. They are not glamorous. They are the floor of any speaker recognition system that actually works.
"Voiceprint will be the best choice for logging into smart gadgets with small screens or without screens." - Lawrence Chen, 2017
SpeakIn's headline number on voiceprint identification, above the 95% industry baseline the company cites.
Started in Shenzhen the year voice was still mostly used to set alarms. Chen was betting earlier than that.
Named to the Forbes 30 Under 30 Asia list for Enterprise Technology in 2018.
Over six in ten SpeakIn employees work in engineering and research roles, mostly on the audio stack.
SpeakIn's Series A, reported at roughly RMB 100 million, was led by IDG Capital. A Series A2 followed, led by Originals Capital.
MIT, Harvard, HKUST and Microsoft Research Asia all appear in the bios of his founding team.
He goes by Lawrence in English and Chen Haoliang at home. Two names, one voiceprint - which is, fittingly, his point.
His company's tech profile lists Nvidia GTX 1080 GPUs and Xeon E5-2697A v4 CPUs. He was doing GPU-accelerated speaker recognition before that became a polite topic at conferences.
SpeakIn's keyword profile reads like a signal processing textbook: Hamming, Hann, Kaiser, Blackman-Harris, Welch, Gaussian, Bartlett. Audio is not a vibe. It is math.
More than 60% of SpeakIn is R&D. This is a founder who treats every salesperson as a downstream consequence of being right about the model.
HQ in Shenzhen. R&D arm in the US. Founder formerly at Google Glass. Customers in Chinese banks and public security. The map is busy.
The MIT dropout line gets used too often. He uses it almost never.
The short version of Chen's thesis: the next decade of consumer hardware will keep eating screens. Earbuds, glasses, watches, ambient devices, the embedded compute inside a car or a kitchen appliance - none of them have room for a six-character password and a forgot-it link. Voice is the only modality that gracefully survives the loss of a display. If a machine is going to take instructions from you and act on them, it has to be sure you are you. That is what SpeakIn does.
The longer version is that biometrics, broadly, are going to keep moving from novelty to default in financial services, and from default in financial services to default in everything. Banks are the first market because banks have the most to lose. Public security is the second because public security has the most to find. Consumer hardware is the third, and it is where the volume lives. Chen has been positioning for that order since 2015. He has been right so far.
Make voice the default login for a world of small-screen and screenless devices. Replace fragile password-based identity with biometric speaker verification in banking, public services, and the next generation of consumer hardware. Hire engineers. Build the model. Let everything else follow.