Exploration of the core functional points of Jiangzhi Robot's human-computer voice interaction technology

Exploration of the core functional points of Jiangzhi Robot's human-computer voice interaction technology

Undoubtedly, the human-machine voice interaction skill of robots is one of the core functions that artificial intelligence robots must have. Some domestic companies, such as iFlytek and Baidu, have been working on human-machine voice technology for many years and have made great progress. Domestic related robot companies, scientific research institutes, etc. have also been constantly trying on this basis in combination with market scenarios. Up to now, there is still no case where human-machine voice interaction skills can support the actual use of market scenarios.

Until the Chat GPT proposed by OPEN AI, which is strongly supported by Microsoft, brought expectations to the world. Chat GPT's current human-computer voice interaction skills have been significantly improved. It can be said to be a revolution. Basically, it can accurately judge the main intention of people to express, and further judge what people may express in the next sentence.

But when we calm down and think further: If we have Chat GPT, can we definitely have good human-computer voice interaction like human-to-human communication? Can we achieve the effect of human-to-human communication?

The answer to you is definitely not yet! The reason is that iFlytek's voice interaction is still Chat GPT, and their main method and main direction are deep learning based on big data training. For the basic aspects of public scenes with certain logical relationships, they will definitely get better and better, and even faster and more accurate than human responses in the future.

But let's think about it carefully: How many of the things we usually communicate are "big things"? It should be more about "family matters" or business dealings, etc. Topics that cannot even be made public. There are also temporary decisions, etc. And the proportion of conversation content in this regard is even greater. At present, Chat GPT skills cannot collect these contents in real time for model training. In other words, Chat GPT's human-computer voice interaction for such content is not the direction and purpose of their research and development.

Through in-depth market analysis and down-to-earth efforts in recent years, we at Jiangzhi Robotics have continuously supplemented the skills needed for voice interaction scenarios that iFlytek and Chat GPT lack. Moreover, these contents can only be completed by users or people they can rely on. We only provide specific operation methods. Moreover, these methods can be completed by anyone with a high school education and basic computer operation skills. At the same time, the content of yesterday's interaction errors can be accurately improved and trained later, so that there will be no mistakes when asking this question tomorrow.

The software we developed based on this method and thinking has gradually shown its advantages in actual interaction effects through more than half a year of user experience testing. With continuous improvement, we will definitely be able to achieve voice interaction in professional scenarios and achieve the effect of human-to-human interaction.

Shenzhen Jiangzhi Industrial Technology Co., Ltd.

March 5, 2023