Components

1. Automatic Speech Recognition (ASR) and Audio Mining

AppTek provides ASR for Modern Standard Arabic (MSA) that is speaker independent. The system supports MSA but is designed to collect, transcribe, and train using selected dialects. The system can support multiple local dialects and improves the accuracy of performance.Spoken Language: the Natural Dialect

• MSA is mostly engineered for broadcast and is rarely used.
• Spontaneous speech is almost never used.
• MSA PlainSpeech includes dialect-adapted modules.

2. Knowledge Management System 

In the area of Knowledge Management, AppTek offers several software modules under the collective name PlainKnowledge, which have already been integrated into third party products or have also been implemented as SDK for end customers. The individual modules are PlainCluster (grouping), PlainClassify (classification), PlainSummarize (summarization), PlainExtract (content-specific extraction), PlainLingua (language recognition of a text), and PlainRetrieve (associative search). 

3. Machine Translation (MT) Technology 

AppTek provides the Machine Translation (MT) application for more than twenty (20) different language pairs. The system can operate on its own or can be used as a dynamic, Web-based translation tool for online content. It is fully integrated with Automatic Speech Recognition and Text-to-Speech software, which enables future development of speech-to-speech Machine Translation. The suite of linguistic tools includes proper-noun recognition, morphological analysis, and large, annotated lexicons.

4. Digital Assets Management (DAM)

The Digital Assets Management system is comprised of multiple agents that provide access to various resources. These resources are collected, stored, and retrieved via the Digital Asset Management Network. The network refers to the interconnection and collaboration of the various agents; it is not a computer-based network. The entire network can be contained within a single computer, but this may not be practical given the resources required to perform real-time analysis. In some cases, agents may exist on the same computer to provide less-latent access to published data. AppTek’s human language technology and DAM have been integrated in the UIMA environment, an open architecture for the management of unstructured information such as natural language text (e.g., news articles), audio, and video. This management includes analysis, search, storage, information retrieval and extraction, data mining, etc. UIMA analyzes unstructured information in order to produce structured databases or indexes. 

5. NameFinder

NameFinder™ is an advanced technology engine that scans text for proper nouns (such as human names, organizations, geographic locations, currencies, parts, etc.) in various languages, and even recognizes proper nouns in writing systems that do not use capitalization. NameFinder™ can also accurately identify the ethno-linguistic origins of a person’s name. The system recognizes and transliterates between the Roman (Latin) alphabet and other writing systems, and can identify transliterated names despite discrepancies, ambiguities, or simple misspellings.

Functionality

Television signal reception and distribution and Telephony Audio Monitoring capability

This solution provides 24x7 monitoring of a foreign language channel, English-speaking channel, or telephony platform. The monitoring capabilities include:

• 24x7 capture of all channel(s)
• Near-real-time transcription with adjustable latency
• Near-real-time machine translation from supported languages and to English of transcribed text for the channels
• Speaker Identification: recognizes voices that match a set of trained speaker models, labeling speaker turns with names. Unspecified speakers are classified according to gender, and clustered together if there are multiple appearances
• An optional  use of a standard Web browser: MediaSphere News Monitoring users interact with the application using a standard browser-based interface. The interface is a complete, easy-to-use system.
• Automatic highlighting of text to mark entities including people, places, organizations, and dates and times. The entity detection and highlighting takes place in two stages:
  • Near-real-time (within limited latency): The optional Name Extraction module identifies names in the speech recognition output, and categorizes them into people, locations, and organizations.
  • Advanced: Post-processing of transcribed text.

Review streaming video with synchronized transcriptions and translations 

Review Streaming Video: The system automatically builds a rich, structured index in real time that can be used to search video content and immediately locate a specific video segment for playback.  The video index is time synchronized to every encoded copy providing immediate and exact retrieval of the content. Words and phrases are automatically highlighted within a transcript, allowing users to monitor the Arabic (source) speech-to-text recognition and English translation synchronized with the video. 

Review Video Segments: The Video Digitizing process transforms standard analog or digital video into Web-friendly and broadcast-quality content. Video Digitizing manages multiple digitizing processes while simultaneously indexing video. VideoCapture automatically detects visual scene changes, spoken words, and recognized voices. The resulting rich video index provides both editors and viewers with fine-grained control over content. The video index allows users to select video segments based on channel, start time, video frame, or between live (with 10-minute latency) or archived videos. 

Review Still Frames: VideoCapture uses signal analysis algorithms to generate key frames, which provide a visual overview of the content. 

Search for specific spoken content in the video cache by Arabic or English-language text query 

Full-text search in Arabic or English: Using Advanced Search, users can narrow searches to include a specific broadcaster, program, language, asset type, date or range of dates, Boolean operators, or keywords. Search queries can be saved and used again and again. The search capability supports all languages managed by the system. The linguistic-based search integrated with the system (TextFinder™) provides the following advanced capabilities:

• A particular feature of this system is its multilingual Morphology Analyzer. The Morphology Analyzer can analyze and generate all inflectional paradigms for all classes of verbs, nouns and adjectives, or any other part of speech. The module is also capable of reducing character strings to their dictionary look-up form (root or stem, as required). The analyzer processes stems as well as meaningful root-derivational morphology.
• Query Translation: TextFinder™ has bilingual dictionaries that allow users to search in multilingual text simultaneously. The user can type a search word or phrase in one language and retrieve the matching words with successive part(s) of speech in other languages using the Morphology Analyzer as well as the large, dynamic, bilingual lexicons.
• Proper Noun Search: A Proper Noun Extraction System is used within TextFinder™. The system can identify all proper nouns in multiple languages such as human names and names of geographic entities such as countries, cities, places, etc., utilizing the large names database developed by AppTek as well as surface parsing and statistics. The search can also be performed using exact- match and fuzzy-match techniques. In addition, this feature allows for Name/Event searches.

 Reference lists and real-time alerts  Alerts from keyword-based filters: MediaSphere monitors news broadcasts and alerts users of new content that is of interest to them. All incoming information is matched against users’ saved queries, and alerts are triggered when the specified criteria in the saved queries matches the incoming information above a specified threshold. The use of contextual profiling and personalization ensures that subtle differences in context are taken into account before alerting a user of new information. It is crucial that any system that alerts a user of new content does so only when the content is relevant to the user and the system allows for an arbitrary level of customization to take place. Embedded proper noun identification is open to user-specific keyword list creation, modification, or addition.  

Provide graphical temporal display of the occurrence of topics or keywords within one or all channels 

The Video Digitizing technology includes the following features:

• Manages, monitors, and schedules video acquisition
• Automatically builds a rich, structured index in real time that can be used to search video content and immediately locate a specific video segment for playback.  The video index produced by VideoCapture is time synchronized to every encoded copy, providing immediate and exact retrieval of content
• Provides automatic alignment of text (e.g., script) with content
• Allows video clips to be played based on a time code associated with the selected key frame or text fragment

Other Functionality

• Administrator Capabilities: The system offers the following capabilities:
  • Web-based interface
  • Index management
  • Account management
  • View management
  • User and group management
  • Security management
  • Folder management

• Automatic Hyperlinking
  • MediaSphere News and Telephony Monitoring automatically generates hyperlinks to relevant material in real time. The hyperlinks connect to contextually similar content and can be used to recommend related articles, documents, affinity products or services, as well as media content that relates to textual content. Because hyperlinks are automatically inserted when a document is returned, they can include references to archived material in addition to the latest news or material on a particular subject.