Samsung Research in Vietnam is a part of a sequence in regards to the individuals and improvements permitting cell AI to reinforce extra lives
Samsung is pioneering premium cell AI experiences. To learn the way Galaxy AI is maximizing the potential of its customers, we’re visiting Samsung Research facilities world wide. Now supporting 16 languages, Galaxy AI is enabling extra individuals to increase their language capabilities, even when offline, due to on-device translation in options comparable to Live Translate, Interpreter, Note Assist and Browsing Assist. We just lately visited Jordan to be taught the complexities of creating an AI mannequin for Arabic, a language with many dialects. This time, we’re going to Vietnam to discover how knowledge is ready to coach AI fashions.
What is the distinction between a ghost, grave and mom in Vietnamese? For a language spoken by 97 million individuals worldwide, little or no. Each phrase interprets to “ma,” “mả” and “má,” respectively — and may solely be distinguished by tone. This illustrates how troublesome it may be for AI fashions to be taught a language, contemplating they can’t acknowledge firsthand the context and feelings of conversations nor the intentions of these talking.
Samsung R&D Institute Vietnam (SRV) used finely refined knowledge to assist its AI mannequin correctly acknowledge even probably the most refined variations in language.
The high quality of knowledge used immediately impacts the accuracy of automated speech recognition (ASR), neural machine translation (NMT) and text-to-speech (TTS) — processes that assist Galaxy AI options comparable to Live Translate, Interpreter, Chat Assist and Browsing Assist break down language limitations.
A Typhoon of Challenges
“Vietnamese is a complex and diverse language with rich expressions, many of which are challenging to capture,” says Ngô Hồng Thái, NMT lead at SRV. Of the 16 languages that Galaxy AI helps, Vietnamese was notably troublesome to develop.
“Personally, creating an AI model for Vietnamese was more daunting than our typhoons!” he provides earlier than explaining the hurdles confronted in the course of the growth course of.
Vietnamese is a tonal language with six distinct tones. As evident within the “ma” instance above, small nuances in vocalization can drastically alter the meanings of phrases. Therefore, a meticulous and detailed method was mandatory.
“When similar sounding words are broken down, one word consists of several short segments, or ‘frame sets’,” says Bui Ngoc Tung, ASR lead at SRV. “The AI model differentiates between the short audio frames of around 20 milliseconds to recognize what words correspond to a certain set of consecutive frames. As such, it is critical to put great effort into the early stages of the AI learning process.”
Furthermore, homophones and homonyms are frequent in Vietnamese. People can usually depend on context and nonverbal parts in conversations to distinguish between phrases that sound the identical or are written the identical however have totally different meanings. However, AI fashions must be taught to precisely establish and differentiate between tones and comparable phrases.
“This isn’t a straightforward task,” Thái explains. “Apart from the amount, the data needs to be accurate to ensure it is capable of recognizing the linguistic nuances that exist in Vietnamese.”
Rigorous Preparation
The knowledge refinement course of consists of three steps. First, the audio and textual content used to coach the AI mannequin should be reviewed and corrected. Then, this dataset goes by means of random checks for total high quality. Finally, the dataset is normalized and cleaned earlier than use in coaching.
“We thoroughly performed a series of tests to check the accuracy of our dataset,” says Nguyen Manh Duy, TTS lead at SRV who oversees database creation. “We confronted numerous surprising issues together with misspelled phrases in scripts and background noise or incorrect pronunciation throughout…
