Court intelligent voice system
Current situation and demand
In the process of court trial, it is necessary to form a written court trial record to facilitate tracing or tracking afterwards. The existing manual dictation methods are limited by the professional proficiency and typing speed of the recorder, resulting in low recording efficiency and waste of manpower. The voice of using information means to improve recording work and improve work efficiency is rising day by day.
In order to promote the modernization of the trial system and trial capacity and accelerate the construction of smart courts, according to the general idea of "wisdom and innovation", Asia News prestige makes full use of intelligent technologies such as cloud computing and big data, focuses on the deep integration of intelligent speech recognition technology and conference scenes, transcribes the whole process of court trial, and effectively manages, analyzes, utilizes and The application of intelligent speech recognition technology can convert the speech into text in real time, and the manuscript can be completed after the court trial, which greatly reduces the requirements for the professional quality of the recorder and reduces the work intensity of the recorder
Court intelligent voice system architecture
During the trial, the voice of the judge, the plaintiff and the defendant was transcribed in real time and pushed to the clerk's computer in real time in the form of text. The clerk can proofread and edit in real time, and the electronic record can be generated after the trial
Using speech synthesis technology, (1) play court discipline. Contents such as pre-trial notice: (2) broadcast relevant legal documents to parties with visual impairment and illiteracy; (3) read out the provisions of laws and regulations, testimony, judgment, etc.: (4) contents that the presiding judge believes need to be broadcast temporarily during the trial
Real time retrieval of laws and regulations, judicial interpretation, case judgment and other information stored in the court knowledge base by voice or text
Automatically distinguish the roles of speakers, such as presiding judge, prosecutor / plaintiff, defendant, etc. When the system transcribes the speech content, the recognition result will automatically correspond to the corresponding role
Custom speech model
For different regions and different types of cases, the system can carry out customized training of speech recognition model in advance. Improve the accuracy of speech recognition.
Real time editing:
During the trial, the clerk can modify, delete and replace the text content output by voice transcription in real time, so as to make the trial record more accurate and refined
Where the error rate of speech recognition effect is high due to noisy scene, fierce quarrel, heavy dialect accent and other factors, and the clerk has no time to modify immediately, you can put an electronic label here. After the trial. The clerk clicks the electronic tag here, and the system will automatically jump to the corresponding video time point to correct the text through the video.
It can carry out real-time voice transcription of various meetings of the court, such as the trial committee, pre-trial preparation meeting, civil mediation, letters and visits, etc., and generate meeting minutes
Advantages of court intelligent voice system
High recognition rate:
The latest generation of tdn-lstm (time delay + time recursive neural network) technology and advanced acoustic model and language are adopted
With the model training method, the speech recognition rate is at the international leading level, and the accuracy of Mandarin speech recognition is more than 95%
Fast recognition speed
The highly optimized decoder core is adopted and the decoding network based on finite state machine (WFST) supports real-time output of speech stream, which greatly improves the speed of speech recognition. The recognition real-time rate is less than 0.3, that is, the recording with audio length of 3 seconds can be processed in 1 second.
The speech recognition model can be customized according to the dialects of specific regions (such as Cantonese), or according to civil, criminal, intellectual property and other trial types, so as to further improve the accuracy of speech recognition
It adopts the most advanced deep neural network and other advanced algorithms to process data by imitating the mechanism of human brain, which greatly improves the recognition rate and operation efficiency
Under the framework of the unified platform, speech recognition, speech synthesis, semantic understanding and handwriting recognition. Image recognition, voiceprint recognition, face recognition, fingerprint recognition and other intelligent modules can be superimposed and combined flexibly, with strong multi service support ability
Provide standard and open AP and SK interfaces, which can directly connect with the existing systems of the court, such as video system and knowledge management system, make full use of existing resources and avoid repeated investment
Stable and reliable
Hardware equipment shall be mature products of high quality. Intelligent capability software is independently developed on the basis of scientific research achievements of Tsinghua University and has been successfully applied in many fields
Improve the informatization level
Artificial intelligence technology is used to build a voice cloud support platform to provide basic support services such as voice transcription, image recognition and big data analysis. On this basis, an auxiliary system for recording and transcription of the trial process is built.
Unstructured data structure
The audio and video content is transferred into text storage by voice transfer technology, and the electronic evidence such as documents and bills are also transferred into text content storage by image recognition technology.
Intelligent platform features
Under the framework of the unified platform, it can flexibly combine speech recognition, speech synthesis, semantic understanding, handwriting recognition, image recognition, voiceprint recognition, face recognition, fingerprint recognition and other capabilities, and has strong multi service support capabilities
Provide a variety of open AP and SDK standards, have good software and hardware compatibility, support multi-channel and a variety of third-party product access processing, support user secondary development, and meet diversified business requirements
It has its own load balancing module, supports cluster deployment, multi-point hot standby, provides real-time monitoring and alarm service, and has good system stability