A former Huawei employee started a business with an AI translation earphone, ranking first globally and achieving an annual revenue of 200 million yuan | Insight Global

Editor's Note: As globalization becomes the core strategy for many Chinese companies, how to conquer the global market has become an extremely professional topic. In the evolution of globalization, many Chinese brands have stood at the forefront. In view of this, Yingke has launched the "Insight Global" column to explore the cutting - edge directions and opportunities of global brands from the growth and changes of brands, providing thinking and inspiration for Chinese brand players and the industry.

This is the 41st issue of our column. In 2016, a former Huawei engineer team targeted the language service track. They overcame the pain points of cross - border communication with a 5 - second ultra - fast translation delay and 90 - decibel noise reduction technology. In 2023, they topped the sales list of translation devices on Amazon in the United States, and their service network covers 170 countries and regions... How can Chinese hardware brands build a technological moat in the red ocean of AI translation with "natural dialogue logic + in - depth exploration of vertical scenarios"? Based on this, we had a chat with Wu Weibing, the CMO of Timekettle.

Text | Lin Qingqing

Editor | Yuan Silai

Tian Li, the founder of Timekettle, decided to start a business after an accident involving his mother.

In 2015, Tian Li's parents decided to travel to Europe. Before departure, Tian Li's mother downloaded a screen - full of translation apps, and Tian Li specially prepared a translator for the old couple.

Abroad, the translator was unexpectedly difficult to use. Even for a simple task like buying a bottle of water, they had to wait for the machine to react for a long time.

When the journey reached a mountain in the French - speaking area of Switzerland, an accident happened. The language problem became a matter of life and death.

Tian Li's mother felt unwell, with a headache and fatigue, presumably due to altitude sickness. The two rushed to the local hospital, only to find that they couldn't communicate at all. Facing the emergency nurse, the old couple spoke into the translator hurriedly, but the translation results were fragmented because of their fast - paced speech. When they had no choice but to gesture desperately with their hands, the nurse was burying her head in studying the translator. Most of the time, they exchanged Chinese and French without understanding a single word of each other.

An anxious language vacuum zone thus emerged at this critical moment.

It took them some time to finally grasp the rhythm of the translator. They had to slow down their speech and enunciate each word clearly to barely communicate.

When Tian Li heard about this from his parents, he felt a chill down his spine in an instant: If his mother had a more serious illness instead of altitude sickness, would irreversible consequences have occurred?

He was no stranger to cross - language communication barriers. Translation apps were often used in some business negotiations during his previous business communications. At that time, he never thought of changing the status quo himself. After leaving Huawei, he went to a robotics company and led hardware development. He thought about starting a business but didn't see any opportunities.

His parents' experience in Europe gave him an inspiration. According to data from the World Health Organization, more than 1 billion people around the world cannot communicate freely due to language barriers.

From handheld translators to mobile apps, traditional solutions have never been able to solve the pain point of "natural dialogue". "No one wants to pass a translator back and forth when arguing," Tian Li once said.

Obviously, this is a vast and unmet market. Tian Li hoped to create a "device that makes people forget about translation", and the in - ear headphones were the closest to his idea.

In 2016, Tian Li found his college classmates and joined forces with several engineers to establish the Timekettle brand and started making wearable translation headphones.

They experienced major product failures and adjusted their directions. It wasn't until 2020 that they launched their first truly successful headphone, the M2. This product quickly sold 100,000 units that year. In 2023, Timekettle topped the sales list of translation devices on Amazon in the United States. In 2024, Timekettle launched the W4Pro series, with a total revenue of over 200 million yuan.

Translation hardware is a traditional product. Overseas, Japan's Post, which currently ranks second in global market share and specializes in the handheld translator Talkmate, is gradually withdrawing from the market due to slow technological iteration (its products have not been updated for three years). Tech giants like Google and Apple are focusing on integrating AI voice assistants with headphone functions (such as real - time translation on Pixel Buds Pro) and have not yet formed large - scale hardware. In China, the market is mainly dominated by translators led by iFlytek.

Timekettle chose the relatively novel model of AI translation headphones and built a technological barrier. On Amazon, they are the only brand merchant besides the cheap translation headphones from Huaqiangbei. So, even though their prices are comparable to those of AirPods, they still have good sales.

Of course, Timekettle's business is far from stable. After ChatGPT initiated the wave of large - scale AI models, the quality and speed of translation have reached a new level. If hardware giants enter the market and integrate, it will directly encroach on the territory of startups.

Ultimate Appeal

Initially, in Tian Li's perception, making translation hardware was a trivial matter that could be solved by simply cooperating with a software outsourcing company.

But as he delved deeper, Tian Li found that to truly achieve cross - language communication, it was necessary to restore the native - language communication scenario as much as possible, and the existing solutions simply couldn't meet the requirements.

"What users want is not translation, but natural dialogue," Wu Weibing, the CMO of Timekettle, told Yingke.

They learned this lesson painfully. In 2017, Timekettle's first translation headphone, the WT2 Plus, was launched on Kickstarter. It raised over one million dollars in crowdfunding with the concept of "real - time translation".

But when geeks put on the headphones for testing, the feedback was embarrassing: The two parties in the conversation had to strictly follow the rhythm of "speak - wait - play". Any interruption would lead to translation chaos. "It's like exchanging secret signals with a robot," an early user complained.

In 2018, they tried a plug - and - play device, but it suffered from poor sales of 200,000 units due to sound - collection defects, resulting in a loss of over ten million yuan and pushing the company to the brink.

This failure made the team completely shift to the headphone category and target two major technological pain points: real - time performance and noise - resistance ability. The serial processing mode of traditional translators (sound collection - translation - playback) causes a delay of at least 2 seconds, and the bandwidth limitation of the Bluetooth protocol further exacerbates data congestion. The turning point came with the "violent disassembly" of the technical path. The Timekettle team split the single - channel Bluetooth into four channels (independent processing of uploading and downloading for both ears), just like forcibly opening four parallel lanes on a single - lane road.

"It's like building a transportation hub inside the headphones," Wu Weibing described. "One small mistake, and the data from the left and right ears will 'collide', turning the translation results into garbled code." It took them 18 months to finally achieve the "translate - while - speaking" function, doubling the conversation efficiency.

Noise resistance was another tough battle. In a test in Shibuya, Tokyo, the environmental noise caused the translation accuracy to plummet to 70%. The Timekettle team developed the RNNVAD voice recognition algorithm, which can filter 90% of background noise through sound - source direction locking and distance perception. "We even asked engineers to wear the headphones and test in nightclubs to ensure accurate sound collection in a 90 - decibel environment," Wu Weibing said. This technology increased the translation accuracy to 93% in busy urban areas, enabling scenarios such as ordering in bars and inquiring at airports.

Technical trade - offs have always been the core challenge in the R & D process. In 2021, in order to reduce the translation delay, the team once simplified the algorithm verification process, which led to a decline in translation accuracy in business scenarios. A cross - border trading company reported that the headphones misinterpreted the "force majeure clause" in the contract as the "powerless clause", posing legal risks.

"Wrong translation is more fatal than waiting," Wu Weibing emphasized at an internal review meeting. After that, the team began to make adjustments. For daily conversations, they call on six major translation engines such as Google and Microsoft to meet general needs, and in vertical fields such as medical and legal, they establish their own specialized term libraries. According to a third - party test report, when it comes to medical terms such as "coronary artery bypass grafting", Timekettle's translation accuracy increased from 76% of the general engines to 92%.

To balance professionalism and usability, they also added a "scenario mode" function - using formal written language in business negotiations and automatically switching to colloquial expressions in travel scenarios. For example, a frequent business traveler gave an example: "'Let’s talk tomorrow' is translated as 'Please discuss it tomorrow' in business mode and 'Let's talk tomorrow' in travel mode."

After the launch of the M2, Timekettle finally climbed out of the trough. They then launched the M3 and W3, which became increasingly popular. In 2023, Timekettle topped the sales list of translation devices on Amazon in the United States.

But at this time, the real battle was just beginning.

Surrounded by Knock - offs, Breaking Through in the B - end Market

There is no electronic product that Huaqiangbei can't copy.

When Timekettle became the number one in the niche category, the white - label manufacturers in Shenzhen's Huaqiangbei naturally took action. They launched "AI translation headphones for $19.9". Some of them used short videos of "sweet conversations between cross - border couples" on TikTok for marketing, and their monthly sales soared to 200,000 units. These products are actually generic headphones paired with open - source translation apps. The actual measured delay is 3 seconds, and the return rate is over 40%.

The entire industry has entered a dark period.

"Knock - offs ruin the perception of the entire category," Wu Weibing said with a bitter smile. What's more troublesome is that the low - price strategy attracted a large number of first - time users. They mistakenly thought that "AI translation should be this cheap", forcing Timekettle to double its budget for market education.

Facing the low - price impact of knock - offs, Timekettle chose a dual - track defense strategy. In the C - end market, the team launched the T1 New Edition priced at 1,299 yuan, retaining the core function of "two - way simultaneous interpretation" but reducing the number of offline languages from 40 to 13. "We're like making the 'youth edition' of mobile phones, cutting out non - essential functions but must maintain the technical bottom line," Wu Weibing explained.

This product ranked among the top three in the sales of translation devices during Amazon Prime Day, successfully intercepting some price - sensitive users.

The B - end market has become a battlefield that Huaqiangbei manufacturers can't reach. Timekettle can indeed present excellent translation indicators: a 5 - second ultra - low delay and 90 - decibel environmental noise reduction - the former is the offline translation delay speed (mobile phones rely on network delay), and the latter is the directional noise - reduction ability to only recognize the speaker's voice (the accuracy of mobile phone microphones in the same environment is less than 50%).

Technical performance has become the key for enterprises to pay a premium. In 2023, Timekettle reached cooperation agreements with Marriott and Hilton, integrating translation headphones into the hotel service system - foreign guests can rent the devices when checking in, and the daily rental fee is only 1/10 of the cost of human translation. At Shenzhen Yingke Law Firm, lawyers even wrote the headphones into the confidentiality agreement: "Clients must use our devices for conversations and are prohibited from using third - party translation software," Wu Weibing recalled.

B - end users are extremely sensitive to the effects, and Timekettle's products have met their expectations. When nurses at Shenzhen Yunshan Medical Center receive foreign patients, the misdiagnosis rate drops by 40% after using their headphones. Tour guides at Shenzhen Airport use the headphones to communicate with overseas local tour operators, and the complaint rate drops by 75%. A church in Mexico City uses the headphones for bilingual services, and believers participate in English Bible discussions for the first time. "Large companies can copy functions, but the know - how of software - hardware integration requires three years of precipitation," Wu Weibing said.

Currently, Timekettle covers 40 languages and 93 accents, with an average accuracy rate of over 95%. It performs particularly well in professional fields such as medical and legal. Nearly 60% of their revenue comes from Amazon, and the rest is evenly split between independent websites and offline channels. Their products cover 170 countries.

Since its establishment, they have witnessed many stories. An American user named Christian once came to thank them. His wife is Venezuelan, and he uses the headphones frequently to communicate with his wife, father - in - law, and mother - in - law. He said that "the product saved his marriage". A Japanese user was also able to accurately describe the symptoms of acute appendicitis to a doctor through the headphones.

In the next three years, industry competition may focus on multi - modal translation (synergy of voice, text, and images) and in - depth exploration of vertical scenarios. The gradual maturity of large - scale AI models also brings more urgency to startups like Timekettle.

In any case, when the competition in hardware parameters becomes extremely fierce, behind Timekettle's breakthrough is still a simple ideal: The ultimate value of technology is to make people truly human again. They don't need to learn language rules or adapt to machine logic. They just need to speak as instinctively as possible.

"After all, we're not just a company that makes headphones. We're here to solve cross - language communication problems," Wu Weibing said.