Page 335 - AI for Good Innovate for Impact
P. 335

AI for Good Innovate for Impact



               (continued)

                Item                Details
                                    Smart assistant has been deployed on the computers of two China Tele-
                                    com business halls. It has also been deployed on PC and pads in the             4.3 - 5G
                Testbeds or Pilot  laboratory which can assist service representatives with product display,
                Deployments         automatic form filling, and SIM card delivery, freeing up 30% of repre-
                                    sentatives' time for high-value tasks such as customer communication
                                    and marketing.

                Code Repositories   Not Available



               2      Use Case Description


               2�1     Description

               For frontline market operations, China Telecom developed the “Hui Xin” LLM (hereinafter
               referred to as Hui Xin), by leveraging its natural advantages in computing power, data, and
               technology. Hui Xin is fine-tuned and distilled using massive volumes of telecom business
               specifications and operational guidelines. It possesses the efficient advantage of Mixture-
               of-Experts (MOE) [4] distributed inference and excels in understanding product marketing,
               business acceptance, and customer service. This drives the digital and intelligent transformation
               of China Telecom's various frontline market channels.
               Telecom service halls are critical hubs where operators provide a wide range of communication
               services to customers. However, in recent years, frontline service representatives in these
               halls face increasing work pressure and declining service efficiency due to complex customer
               demands, surging transaction volumes, cumbersome system operations, high staff turnover
               rates, and a large influx of new employees. There is an urgent need to implement artificial
               intelligence (AI) technologies to assist these representatives in consistently delivering high-
               quality customer service.

               To address this, China Telecom, based on the Hui Xin, has developed a smart assistant for
               telecom business acceptance [1]. This smart assistant breaks through the limitations of traditional
               Graphical User Interface (GUI)-based acceptance mode, creating a novel Language User
               Interface (LUI) acceptance paradigm. It provides service representatives with comprehensive
               AI auxiliary functions, empowering all service representatives to achieve zero onboarding
               training, gold-standard service quality and hands-free system operations, greatly reducing
               their work pressure and improving work efficiency. 

               The Hui Xin-based smart assistant effectively acts as an AI-powered secretary for service
               representatives. It monitors, captures, and processes the real-time face-to-face dialogue
               between the representative and the customer. The smart assistant provides the representative
               with:

               •    End-to-end processing of real-time conversations: When a customer enters the telecom
                    service hall, the service representative initiates the interaction with a greeting and inquiry
                    about needs. A speech recognition model instantly detects the conversation between
                    the service representative and the customer [3]. The speech stream first goes through a
                    speech enhancement model to reduce noise and improve voice quality. Then, a speaker
                    diarization and recognition model separates the speech of the service representative
                    and the customer and identifies which utterances belong to each. Subsequently, a



                                                                                                    299
   330   331   332   333   334   335   336   337   338   339   340