TIN

TIN is a system for fast multi-criterion text phrase search in larger audio and video databases. Full-text indexing is performed on recordings’ transcripts obtained through speech recognition. It is a server-client solution for Windows which exhibits high scalability with web client applications.

TIN has been designed mainly for use by internal security agencies. It enables full-text search of recordings database required for operational and procedural activities. Large database search performed manually can be very time and resource consuming, and therefore expensive. Full-text indexing of the transcripts can speed-up that process, considerably decrease its costs and enable new ways of processing and use of the database.

The main applications of the system is a text search of the following types of databases:

• emergency phone system recordings
• phone info line recordings
• operational recordings

Other applications include text search of conference recordings, parliament sessions and court proceedings.

TIN performs automatic speech recognition and full-text indexing of the obtained recordings’ transcripts. It can process a large number of recordings in parallel thanks to its scalability. TIN exploits the original ASR engine adapted to process recordings of various types. Various engine settings dedicated to these types are defined which facilitates speech recognition optimization.

TIN performs text phrase search in various modes, e.g. verbatim as specified by a user or including all grammar forms of the specified words. The search process can utilize standard thesaurus as well as synonym sets defined by the user. The latter feature allows one to adjust the search process to specific vocabulary used in the recordings.

TIN system scheme

System ensures accountability and non-repudiation through:

• user authentication – TIN can be integrated with an external authentication service
• authorization – system administrator defines a set of operations each user can perform on various recording sets
• key user actions are logged

TIN is a server-client solution with the following features:

• a web application is used to access recordings and search results
• data (recordings, their metadata and transcripts) can be exported and searched locally on the user’s machine (in case user does not have direct access to TIN)
• TIN does not require communication with any external service, i.e. service outside of the local network

TIN encompasses the following components:

• Recording Repository offering reliable storage
• Database for recordings metadata and transcripts as well as user data
• User authentication service (optional)
• TIN module supervising speech recognition and text indexing
• ASR engine
• User applications
  • system management web application
  • recordings management desktop application
  • recordings search and browsing web application

TIN is one of the results of ARM-1 project. It was designed in close cooperation with internal security agencies. TIN is a server solution for Windows.