Core Tip: In the era of further heated development of AI, the division of the AI industry structure is also becoming more refined. Voice recognition technology is a major system of AI technology, and in recent years, with the refinement of definitions and industrial division of labor, “semantic recognition,” which is often confused with “voice recognition,” is highlighting its value. Although voice and semantics differ by just one word, a small deviation can lead to significant errors.
To better understand the differences between semantic recognition and voice recognition, we can vividly express them using human organs: voice technology is equivalent to a person’s mouth and ears, responsible for expression and acquisition, while semantic technology is akin to a person’s brain, responsible for thinking and information processing. Let’s illustrate this with a common product form:
For example, in an in-car system, the interaction between the vehicle and the driver involves the vehicle acquiring the driver’s voice and broadcasting road conditions, which falls under the category of voice recognition. However, how the acquired voice is understood and how the path is planned is another system altogether.
Just like having excellent hearing does not imply that one’s brain is smart. After recognizing the same voice, different machines might respond differently, which highlights the distinction in machines’ understanding of semantics. Just because a student may not be particularly capable does not mean that their speaking and listening skills are deficient; it merely indicates that their processing ability is not as strong, which often depends on their brain.
When smart home technology reaches a certain level, we can sit in front of the TV and use voice commands to control what program to watch. The precision of interaction requires a high level of semantic understanding capability. For instance, when you want to watch the British drama “Sherlock,” it is often mistakenly referred to as “Charlotte the Detective” because “Charlotte” is more commonly used. If semantic understanding is not applied, the search result might be “Charlotte’s Worries,” which is also frequently referenced.
Qi Chao, CTO of Triangular Beast Company, which specializes in semantic recognition, explained the above phenomenon to us: When you cannot remember the full name of the show, semantic understanding needs to correct you and provide a more suitable response. In fact, there is a significant demand from users when watching TV; when users don’t know what to watch, they need machines to assist in recommendations and guidance, and the precision and humanity of this process depend on the level of intelligence.
Qi Chao vividly compared semantic understanding to the process of cooking. The first step of buying ingredients corresponds to data acquisition, the second step of washing vegetables represents data cleaning, and the third step of cooking is akin to the machine learning process, which requires various learning tools, just as cooking requires various utensils and seasonings; machine learning also needs the ability to learn, like having all the essentials but lacking good cooking skills. The fourth step, the implementation of artificial intelligence products, is like the final dish being plated. Each step requires breakthroughs and refinement.
Source: China Smart City Network