The Jun'er Translator is an artificial intelligence translator equipped with a touch screen that can directly calibrate content. Through the online upgrade of OTA, Zhuner Translation Machine supports oral translation in 31 languages, subdivided into accents in 44 countries, and launched a multi-machine multi-language real-time group chat function. Beijing Fenyin Tower Technology Co., Ltd., abbreviated as BABEL Technology, was established by a team of Tsinghua University and a successful serial entrepreneur, focusing on the field of speech language artificial intelligence. Core technologies: voice recognition, voiceprint recognition, machine translation, face recognition.

The voice recognition engine developed by the dial tower adopts leading streaming end-to-end speech language modeling method and integrates independent natural language processing technology. The speech recognition accuracy has reached practical standards, providing a variety of public clouds, private clouds, and offline. Application scenario.

Voiceprint recognition is a type of biometric technology, also known as speaker recognition, which is a technology to determine the identity of the speaker through voice. Relying on optimization measures such as independent algorithm iteration and model upgrade, Dietone Tower Technology has made certain breakthroughs in identifying the speaker's acoustic characteristics such as formant peak, fundamental pitch, cepspectrum, and initially solved the problems of identification accuracy and stability. At present, voiceprint recognition technology has supported the second-level search and recognition of voiceprint libraries with capacity of more than tens of millions.

Machine translation is the ability to use computer technology to translate one language into another. The NMT translation engine independently developed by Dione Tower Technology supports real-time mutual translation between mainstream languages. Provides a variety of application scenarios such as public cloud, private cloud and offline.

Based on the speech information decomposition method, sound features are extracted with good robustness, and specific sensitive words, abnormal sounds, etc. are detected and identified, which are applied to the field of sound security. At present, combined with sparse learning and noise-added training, under strong noise and strong reverb conditions, the EER has an average relative decrease of 10% to 15% compared with the mainstream baseline, and the error rate has decreased by 70% compared to the mainstream system. .