Creating The Right Transcription Engine for Call Centers
Updated: Mar 17, 2020

Creating The Right Transcription Engine for Call Centers
AsiaInfo Product Marketing
Mar 14th, 2020
At the beginning of March this year, the NPC and CPPCC brought the public attention to reporters '"interview artifacts", quickly converted the interview voice content into text, and improved the efficiency of journalists' posting. Dingtalk, it automatically translates the voice information into text. While it is convenient for the sender, it also saves the recipient's time and cost and optimizes the user experiences. It is not an exception to use the transcribing ability to improve efficiency and optimize the experience. In the gigantic voice data storage, the call center's demand for voice transcribing is increasingly prominent. AsiaInfo Speech Translation (Automatic Speech Recognition, ASR) engine is designed to provide speech translation services for enterprise call centers, helping enterprises reduce costs and improve efficiency, and reveal the value of data.
The Speech Rewriting (ASR) engine is an AI-capable tool that translates the voice files entered by customers and returns text results to customers. The voice transcription engine for the call center originates from customer expectations.
In 2018, our customers successively requested AsiaInfo for assistance, and their needs were very simple-phone voice transliteration. Most of these customers have complex and diverse business scenarios and have their own call centers or business systems. They want to analyze the texts themselves based on the actual needs of their business. Based on the understanding of AsiaInfo Veris Virtual Contact Center (Veris VCC) platform, many customers know that AsiaInfo has the transcribing service capability, so they hope to be able to connect with the transcribing engine, help them complete the transcribing, and do text analysis through the business system.
In recent years, AsiaInfo has penetrated into industry scenarios, collected customer needs, and continuously explored the application of intelligent voice transcribing in call centers. After technical precipitation and accumulation, the commercial practice of voice transcribing has exceeded 35 million minutes. Through the continuous evolution and optimization of technology and products, AsiaInfo's voice transcription engine has greater concurrency, faster transcription speed, and more flexibility, and the production capacity of the industry transcription model is continuously strengthened. Based on this, AsiaInfo Transcription Engine is fully open.
Speech Transcription
The customer's demand for speech transcription is mainly reflected in three aspects: the accuracy of speech transcription recognition, flexibility as needed, and cost control.
The accuracy of speech transcription recognition is the basis of quality inspection and analysis. Speech is unstructured data and cannot be retrieved and analyzed. All transcribing has to do is to transform unstructured data such as speech into structured data, which is text information. Text information can be retrieved and keywords extracted for analysis. However, if the transcription recognition is not accurate enough, even if the transcription is completed, it will affect the effectiveness of data analysis. Therefore, the accuracy of transcription recognition must be high.
The transcription engine required by customers also needs to be flexible on demand to meet business requirements. For example, the customer may transliterate 70 hours of voice files on the first day, but the business volume surges the next day, and the transcribing volume will increase from 70 hours to 400 hours. This requires the transcribing engine to stay at a consistent level in response speed and transcribing efficiency under any influence, the capacity can be flexibly expanded.
Cost Control.
If the enterprise call center has 100 hours of recordings to be transliterated, the traditional manual listening recording method requires a dedicated team. But now, only through AsiaInfo's open transcribing engine, voice can be automatically transliterated directly into text, which saves a lot of labor costs.
The AsiaInfo New Technology Lab in Beijing is full of R&D personnel and a large number of experts assisting machine learning. The laboratory cooperates with first-class AI technology companies such as Alibaba or iFlyTech. Based on the underlying artificial intelligence technology and combining customer voice files for model training, the lab builds a transcription engine that focuses on the call center industry scene. To use the transcribing engine, the customer only needs to initiate a transcribing task through the interface. After the transcription is completed, the customer can receive the transcription through a callback, which is simple, efficient and easy to operate.
AsiaInfo Speech Transcription Engine is designed for call centers. At present, the mainstream voice transcription service providers in the market have different focuses. Although the underlying technology has gradually matured, most of them are general-purpose transcription models. For companies with industry attributes, the transcription recognition rate is slowly improved, and the text utility is relatively low.
Based on the service experience of more than 10 years of rooted call centers, AsiaInfo is well versed in the business scenarios of call centers based on “call volume". Adopt privatized deployment to ensure the security of customer data. Based on massive voice data, long-term optimization of customer-specific words and industry terms is provided to serve enterprises with a dedicated voice transliteration training model to improve recognition rate.
Dual-track recording, telephone voice sampling at 8KHz, caller's emotions, and natural language communication are the characteristics of call center business scenarios. Based on these characteristics, develop a transcribing engine that meets the actual needs of the industry, including transcribing text to distinguish speakers, and for telephone voice sampling, the transcription recognition rate, sorting out the words that fit the scene, and training the machine to improve the practical effectiveness of the voice-to-text conversion of the enterprise call center. At present, it not only supports the transcribing of call recordings generated by the AsiaInfo series products, but also supports the voice transcription of third-party call centers.
Since its launch, AsiaInfo's speech transcription engine has covered multiple industries such as insurance, corporate services and new retail, and has provided efficient speech transcription services for a large scale of loyal customer base of AsiaInfo.