[Ming Pao Daily News] Hong Kong-based Zyetric Technologies Limited, which has developed ‘facial translation’, has recently participated in a number of mainland exhibitions, such as the China International Import Expo in Shanghai and the Shenzhen High-Tech Fair, and has moved into the Foshan Hong Kong Macau Youth Entrepreneurship Center as an incubatee. ‘Facial Translation’ refers to the use of AI (Artificial Intelligence) to translate the dialogues and voices of the characters in the screen into a foreign language, while matching lip movements and facial expressions. Company founder Patrick Lam said that the non-real-time facial translation product for films and TV dramas will be launched in December, while the real-time facial translation product for mainland cross-border e-commerce ‘live streaming’ will be launched in the third quarter of 2025, and by participating in trade fairs and exhibitions and setting up business in the bay area, the company can effectively reach their target audience .
Mingpao Reporter Lin Ying
Non-real time service to be released within this year, supports the upload of movies and advertisements.
“I spent half of my time in Hong Kong and half of my time in the mainland” says Patrick Lam, co-founder of Zyetric Technologies Limited. The company’s facial translation product Zhowcase is under final testing and editing, “customers have waited for long, and have constantly asked when it will be released”. Facial translation is a service that uses AI to translate the dialogue of a film character, and at the same time modifies the lip movement and facial expression according to the tone and mood, so that it appears as if the character is speaking in a foreign language (see page A6 of this newspaper on 3 June 2024). Zhowcase can translate between 32 languages and includes both non-real-time and real-time services. It is expected that the non-real-time product will be launched in December, i.e. users will upload movies, TV dramas or advertisement clips, and the AI will generate a version in a different language; and the real-time livestream translation product will be launched in the third quarter of 2025, targeting online celebrities (KOLs) who bring goods for cross-border e-commerce merchants.
“Trying for gold” in Hong Kong, small KOL market required expanding to other markets
Lam explains that he sees KOLs as a target group because of the booming development of cross-border e-commerce in the Mainland and the ‘immediate need’ to introduce products in different foreign languages. He recalls that Zyetric’s first product is Zyviz.com, a service that intelligently inserts in-video advertsements into YouTube clips. ‘The response in Hong Kong was very good. We done more than 400 adverts in the whole of last year’. Zyviz has also accumulated a lot of experience and data for Zyetric, and now there are a number of YouTube celebrities signed to the company. However, Lam believes that the KOL market in Hong Kong is too small, ‘our revenue was only a few tens of millions last year, so we must develop the mainland and overseas markets.
Patrick said, Zyviz.com has thought of co-operating with mainland short film platforms, but because he is not familiar with mainland platforms, has not found a suitable partner, and does not have enough capital, he prefers to use “Zhowcase” to enter the mainland market, “If you sell goods speaking Putonghua, you can reach 200 million audiences, but if it can be translated into 32 languages, it could potentially reach 3 to 4 billion people, and the chance of exporting the products will be much greater”. Lam said that he is well aware of the policies and development trends of cross-border e-commerce in China, “Looking at the figures, I would think that the growth is slowing down or reaching a bottleneck, which can be broken through by our technology.”
Current 40-second delay hoped to be reduced by half, plans to charge ¥100 per 2 hours
Currently, there is a 40-second delay in the real-time translation of live broadcasts. Lam explains that translation requires listening to the whole sentence, and even real human translators have a delay of at least 10 seconds, so Zyetric’s goal is to compress the delay to 20 seconds, but at the same time, take into account costs. ‘Right now I'm using a 2017 GPU (Graphics Processor). If I replace it with a new GPU, it may be 10 times faster, but it's more than 20 times more expensive. Lam emphasised that the real-time service plan of Zhowcase is priced at 100 yuan (RMB) per two hours, ‘Low cost, affordable, and a large user base is the only way to be profitable. The service will be launched in 2025, and by 2027, if there are one million users, each using RMB 100 per month, the company will have a monthly income of RMB 100 million.
Apart from delay, another thing to be improved is the accuracy, which Lam describes as 95% at present, including the accuracy of the translation of sentences and the ‘right lip movement’, “from 95% to 100% is the most difficult part. Now the audience can definitely hear it, but they can tell that it's not the original voice, it's made by the AI. For example, if the clip is translated into Italian, local friends will find it unnatural because Italians don't speak like that. He also said that another reason why he chose to focus on the KOL live-commerce market is that the audience does not have a high demand for accuracy. “If the lip movement is a little off, the audience won't pay too much attention to I, but this is not the case for movies and TV dramas, they have to look good and be accurate, and they can’t look blurry at all.”. Currently, Zyetric is undergoing Pre-A stage funding, with a valuation of ¥75 million and a funding target of ¥15 million, which will be used for continuous product development and quality enhancement, as well as expanding its business in the Mainland.
Article source：MingPao